| Title: | Build Machine Learning Models Like Using Python's Scikit-Learn Library in R |
|---|---|
| Description: | The idea is to provide a standard interface to users who use both R and Python for building machine learning models. This package provides a scikit-learn's fit, predict interface to train machine learning models in R. |
| Authors: | Manish Saraswat [aut, cre] |
| Maintainer: | Manish Saraswat <[email protected]> |
| License: | GPL-3 | file LICENSE |
| Version: | 0.5.7 |
| Built: | 2024-09-15 04:17:30 UTC |
| Source: | https://github.com/saraswatmks/superml |
BM25 stands for Best Matching 25. It is widely using for ranking documents and a preferred method than TF*IDF scores. It is used to find the similar documents from a corpus, given a new document. It is popularly used in information retrieval systems. This implementation is based on c++ functions hence quite optimised as well.
bm_25(document, corpus, top_n)bm_25(document, corpus, top_n)
document |
a string for which to find similar documents |
corpus |
a vector of strings against which document is to be matched |
top_n |
top n similar documents to find |
a vector containing similar documents and their scores
docs <- c("chimpanzees are found in jungle", "chimps are jungle animals", "Mercedes automobiles are best", "merc is made in germany", "chimps are intelligent animals") sentence <- "automobiles are" s <- bm_25(document=sentence, corpus=docs, top_n=2)docs <- c("chimpanzees are found in jungle", "chimps are jungle animals", "Mercedes automobiles are best", "merc is made in germany", "chimps are intelligent animals") sentence <- "automobiles are" s <- bm_25(document=sentence, corpus=docs, top_n=2)
Training Dataset used for classification examples. This is classic titanic dataset used to predict if a passenger will survive or not in titanic ship disaster.
cla_traincla_train
An object of class data.table (inherits from data.frame) with 891 rows and 12 columns.
https://www.kaggle.com/c/titanic/data
Handy function to calculate count of values given in a list or vector
Counter(data, sort = TRUE, decreasing = FALSE)Counter(data, sort = TRUE, decreasing = FALSE)
data |
should be a vector or list of input values |
sort |
a logical value, to sort the result or not |
decreasing |
a logical value, the order of sorting to be followed |
count of values in a list
d <- list(c('i','am','bad'),c('you','are','also','bad')) counts <- Counter(d, sort=TRUE, decreasing=TRUE)d <- list(c('i','am','bad'),c('you','are','also','bad')) counts <- Counter(d, sort=TRUE, decreasing=TRUE)
Creates CountVectorizer Model.
Given a list of text, it generates a bag of words model and returns a sparse matrix consisting of token counts.
sentencesa list containing sentences
max_dfWhen building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold, value lies between 0 and 1.
min_dfWhen building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold, value lies between 0 and 1.
max_featuresBuild a vocabulary that only consider the top max_features ordered by term frequency across the corpus.
ngram_rangeThe lower and upper boundary of the range of n-values for different word n-grams or char n-grams to be extracted. All values of n such such that min_n <= n <= max_n will be used. For example an ngram_range of c(1, 1) means only unigrams, c(1, 2) means unigrams and bigrams, and c(2, 2) means only bigrams.
splitsplitting criteria for strings, default: " "
lowercaseconvert all characters to lowercase before tokenizing
regexregex expression to use for text cleaning.
remove_stopwordsa list of stopwords to use, by default it uses its inbuilt list of standard stopwords
modelinternal attribute which stores the count model
new()
CountVectorizer$new( min_df, max_df, max_features, ngram_range, regex, remove_stopwords, split, lowercase )
min_dfnumeric, When building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold, value lies between 0 and 1.
max_dfnumeric, When building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold, value lies between 0 and 1.
max_featuresinteger, Build a vocabulary that only consider the top max_features ordered by term frequency across the corpus.
ngram_rangevector, The lower and upper boundary of the range of n-values for different word n-grams or char n-grams to be extracted. All values of n such such that min_n <= n <= max_n will be used. For example an ngram_range of c(1, 1) means only unigrams, c(1, 2) means unigrams and bigrams, and c(2, 2) means only bigrams.
regexcharacter, regex expression to use for text cleaning.
remove_stopwordslist, a list of stopwords to use, by default it uses its inbuilt list of standard english stopwords
splitcharacter, splitting criteria for strings, default: " "
lowercaselogical, convert all characters to lowercase before tokenizing, default: TRUE
Create a new 'CountVectorizer' object.
A 'CountVectorizer' object.
cv = CountVectorizer$new(min_df=0.1)
fit()
CountVectorizer$fit(sentences)
sentencesa list of text sentences
Fits the countvectorizer model on sentences
NULL
sents = c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
cv = CountVectorizer$new(min_df=0.1)
cv$fit(sents)
fit_transform()
CountVectorizer$fit_transform(sentences)
sentencesa list of text sentences
Fits the countvectorizer model and returns a sparse matrix of count of tokens
a sparse matrix containing count of tokens in each given sentence
sents = c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
cv <- CountVectorizer$new(min_df=0.1)
cv_count_matrix <- cv$fit_transform(sents)
transform()
CountVectorizer$transform(sentences)
sentencesa list of new text sentences
Returns a matrix of count of tokens
a sparse matrix containing count of tokens in each given sentence
sents = c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
new_sents <- c("dark at night",'mothers day')
cv = CountVectorizer$new(min_df=0.1)
cv$fit(sents)
cv_count_matrix <- cv$transform(new_sents)
clone()
The objects of this class are cloneable with this method.
CountVectorizer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `CountVectorizer$new` ## ------------------------------------------------ cv = CountVectorizer$new(min_df=0.1) ## ------------------------------------------------ ## Method `CountVectorizer$fit` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') cv = CountVectorizer$new(min_df=0.1) cv$fit(sents) ## ------------------------------------------------ ## Method `CountVectorizer$fit_transform` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') cv <- CountVectorizer$new(min_df=0.1) cv_count_matrix <- cv$fit_transform(sents) ## ------------------------------------------------ ## Method `CountVectorizer$transform` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') new_sents <- c("dark at night",'mothers day') cv = CountVectorizer$new(min_df=0.1) cv$fit(sents) cv_count_matrix <- cv$transform(new_sents)## ------------------------------------------------ ## Method `CountVectorizer$new` ## ------------------------------------------------ cv = CountVectorizer$new(min_df=0.1) ## ------------------------------------------------ ## Method `CountVectorizer$fit` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') cv = CountVectorizer$new(min_df=0.1) cv$fit(sents) ## ------------------------------------------------ ## Method `CountVectorizer$fit_transform` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') cv <- CountVectorizer$new(min_df=0.1) cv_count_matrix <- cv$fit_transform(sents) ## ------------------------------------------------ ## Method `CountVectorizer$transform` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') new_sents <- c("dark at night",'mothers day') cv = CountVectorizer$new(min_df=0.1) cv$fit(sents) cv_count_matrix <- cv$transform(new_sents)
Computes the dot product between two given vectors.
dot(a, b, norm = TRUE)dot(a, b, norm = TRUE)
a |
numeric vector |
b |
numeric vector |
norm |
logical, compute normalised dot product, default=True |
numeric vector containing sdot product score
a <- runif(5) b <- runif(5) s <- dot(a, b)a <- runif(5) b <- runif(5) s <- dot(a, b)
Computes the dot product between a vector and a given matrix. The vector returned has a dot product similarity value for each row in the matrix.
dotmat(a, b, norm = TRUE)dotmat(a, b, norm = TRUE)
a |
numeric vector |
b |
numeric matrix |
norm |
logical, compute normalised dot product, default=True |
numeric vector containing dot product scores
Runs grid search cross validation scheme to find best model training parameters.
Grid search CV is used to train a machine learning model with multiple combinations of training hyper parameters and finds the best combination of parameters which optimizes the evaluation metric. It creates an exhaustive set of hyperparameter combinations and train model on each combination.
trainersuperml trainer object, could be either XGBTrainer, RFTrainer, NBTrainer etc.
parametersa list of parameters to tune
n_foldsnumber of folds to use to split the train data
scoringscoring metric used to evaluate the best model, multiple values can be provided. currently supports: auc, accuracy, mse, rmse, logloss, mae, f1, precision, recall
evaluation_scoresparameter for internal use
new()
GridSearchCV$new(trainer = NA, parameters = NA, n_folds = NA, scoring = NA)
trainersuperml trainer object, could be either XGBTrainer, RFTrainer, NBTrainer etc.
parameterslist, a list of parameters to tune
n_foldsinteger, number of folds to use to split the train data
scoringcharacter, scoring metric used to evaluate the best model, multiple values can be provided. currently supports: auc, accuracy, mse, rmse, logloss, mae, f1, precision, recall
Create a new 'GridSearchCV' object.
A 'GridSearchCV' object.
rf <- RFTrainer$new()
gst <-GridSearchCV$new(trainer = rf,
parameters = list(n_estimators = c(100),
max_depth = c(5,2,10)),
n_folds = 3,
scoring = c('accuracy','auc'))
fit()
GridSearchCV$fit(X, y)
Xdata.frame or data.table
ycharacter, name of target variable
Trains the model using grid search
NULL
rf <- RFTrainer$new()
gst <-GridSearchCV$new(trainer = rf,
parameters = list(n_estimators = c(100),
max_depth = c(5,2,10)),
n_folds = 3,
scoring = c('accuracy','auc'))
data("iris")
gst$fit(iris, "Species")
best_iteration()
GridSearchCV$best_iteration(metric = NULL)
metriccharacter, which metric to use for evaluation
Returns the best parameters
a list of best parameters
rf <- RFTrainer$new()
gst <-GridSearchCV$new(trainer = rf,
parameters = list(n_estimators = c(100),
max_depth = c(5,2,10)),
n_folds = 3,
scoring = c('accuracy','auc'))
data("iris")
gst$fit(iris, "Species")
gst$best_iteration()
clone()
The objects of this class are cloneable with this method.
GridSearchCV$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `GridSearchCV$new` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) ## ------------------------------------------------ ## Method `GridSearchCV$fit` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) data("iris") gst$fit(iris, "Species") ## ------------------------------------------------ ## Method `GridSearchCV$best_iteration` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) data("iris") gst$fit(iris, "Species") gst$best_iteration()## ------------------------------------------------ ## Method `GridSearchCV$new` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) ## ------------------------------------------------ ## Method `GridSearchCV$fit` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) data("iris") gst$fit(iris, "Species") ## ------------------------------------------------ ## Method `GridSearchCV$best_iteration` ## ------------------------------------------------ rf <- RFTrainer$new() gst <-GridSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100), max_depth = c(5,2,10)), n_folds = 3, scoring = c('accuracy','auc')) data("iris") gst$fit(iris, "Species") gst$best_iteration()
Calculates out-of-fold mean features (also known as target encoding) for train and test data. This strategy is widely used to avoid overfitting or causing leakage while creating features using the target variable. This method is experimental. If the results you get are unexpected, please report them in github issues.
kFoldMean(train_df, test_df, colname, target, n_fold = 5, seed = 42)kFoldMean(train_df, test_df, colname, target, n_fold = 5, seed = 42)
train_df |
train dataset |
test_df |
test dataset |
colname |
name of categorical column |
target |
the target or dependent variable, should be a string. |
n_fold |
the number of folds to use for doing kfold computation, default=5 |
seed |
the seed value, to ensure reproducibility, it could be any positive value, default=42 |
a train and test data table with out-of-fold mean value of the target for the given categorical variable
train <- data.frame(region=c('del','csk','rcb','del','csk','pune','guj','del'), win = c(0,1,1,0,0,0,0,1)) test <- data.frame(region=c('rcb','csk','rcb','del','guj','pune','csk','kol')) train_result <- kFoldMean(train_df = train, test_df = test, colname = 'region', target = 'win', seed = 1220)$train test_result <- kFoldMean(train_df = train, test_df = test, colname = 'region', target = 'win', seed = 1220)$testtrain <- data.frame(region=c('del','csk','rcb','del','csk','pune','guj','del'), win = c(0,1,1,0,0,0,0,1)) test <- data.frame(region=c('rcb','csk','rcb','del','guj','pune','csk','kol')) train_result <- kFoldMean(train_df = train, test_df = test, colname = 'region', target = 'win', seed = 1220)$train test_result <- kFoldMean(train_df = train, test_df = test, colname = 'region', target = 'win', seed = 1220)$test
Trains a k-means machine learning model in R
Trains a unsupervised K-Means clustering algorithm. It borrows mini-batch k-means function from ClusterR package written in c++, hence it is quite fast.
clustersthe number of clusters
batch_sizethe size of the mini batches
num_initnumber of times the algorithm will be run with different centroid seeds
max_itersthe maximum number of clustering iterations
init_fractionpercentage of data to use for the initialization centroids (applies if initializer is kmeans++ or optimal_init). Should be a float number between 0.0 and 1.0.
initializerthe method of initialization. One of, optimal_init, quantile_init, kmeans++ and random.
early_stop_itercontinue that many iterations after calculation of the best within-cluster-sum-ofsquared-error
verboseeither TRUE or FALSE, indicating whether progress is printed during clustering
centroidsa matrix of initial cluster centroids. The rows of the CENTROIDS matrix should be equal to the number of clusters and the columns should be equal to the columns of the data
tola float number. If, in case of an iteration (iteration > 1 and iteration < max_iters) "tol" is greater than the squared norm of the centroids, then kmeans has converged
tol_optimal_inittolerance value for the ’optimal_init’ initializer. The higher this value is, the far appart from each other the centroids are.
seedinteger value for random number generator (RNG)
modeluse for internal purpose
max_clusterseither a numeric value, a contiguous or non-continguous numeric vector specifying the cluster search space
new()
KMeansTrainer$new( clusters, batch_size = 10, num_init = 1, max_iters = 100, init_fraction = 1, initializer = "kmeans++", early_stop_iter = 10, verbose = FALSE, centroids = NULL, tol = 1e-04, tol_optimal_init = 0.3, seed = 1, max_clusters = NA )
clustersnumeric, When building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold, value lies between 0 and 1.
batch_sizenuemric, When building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold, value lies between 0 and 1.
num_initinteger, use top features sorted by count to be used in bag of words matrix.
max_iterscharacter, regex expression to use for text cleaning.
init_fractionlist, a list of stopwords to use, by default it uses its inbuilt list of standard stopwords
initializercharacter, splitting criteria for strings, default: " "
early_stop_itercontinue that many iterations after calculation of the best within-cluster-sum-ofsquared-error
verboseeither TRUE or FALSE, indicating whether progress is printed during clustering
centroidsa matrix of initial cluster centroids. The rows of the CENTROIDS matrix should be equal to the number of clusters and the columns should be equal to the columns of the data
tola float number. If, in case of an iteration (iteration > 1 and iteration < max_iters) "tol" is greater than the squared norm of the centroids, then kmeans has converged
tol_optimal_inittolerance value for the ’optimal_init’ initializer. The higher this value is, the far appart from each other the centroids are.
seedinteger value for random number generator (RNG)
max_clusterseither a numeric value, a contiguous or non-continguous numeric vector specifying the cluster search space
Create a new 'KMeansTrainer' object.
A 'KMeansTrainer' object.
data <- rbind(replicate(20, rnorm(1e4, 2)),
replicate(20, rnorm(1e4, -1)),
replicate(20, rnorm(1e4, 5)))
km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6)
fit()
KMeansTrainer$fit(X, y = NULL, find_optimal = FALSE)
Xdata.frame or matrix containing features
yNULL only kept here for superml's standard way
find_optimallogical, to find the optimal clusters automatically
Trains the KMeansTrainer model
NULL
data <- rbind(replicate(20, rnorm(1e4, 2)),
replicate(20, rnorm(1e4, -1)),
replicate(20, rnorm(1e4, 5)))
km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6)
km_model$fit(data, find_optimal = FALSE)
predict()
KMeansTrainer$predict(X)
Xdata.frame or matrix
Returns the prediction on test data
a vector of predictions
data <- rbind(replicate(20, rnorm(1e4, 2)),
replicate(20, rnorm(1e4, -1)),
replicate(20, rnorm(1e4, 5)))
km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6)
km_model$fit(data, find_optimal = FALSE)
predictions <- km_model$predict(data)
clone()
The objects of this class are cloneable with this method.
KMeansTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `KMeansTrainer$new` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) ## ------------------------------------------------ ## Method `KMeansTrainer$fit` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) km_model$fit(data, find_optimal = FALSE) ## ------------------------------------------------ ## Method `KMeansTrainer$predict` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) km_model$fit(data, find_optimal = FALSE) predictions <- km_model$predict(data)## ------------------------------------------------ ## Method `KMeansTrainer$new` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) ## ------------------------------------------------ ## Method `KMeansTrainer$fit` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) km_model$fit(data, find_optimal = FALSE) ## ------------------------------------------------ ## Method `KMeansTrainer$predict` ## ------------------------------------------------ data <- rbind(replicate(20, rnorm(1e4, 2)), replicate(20, rnorm(1e4, -1)), replicate(20, rnorm(1e4, 5))) km_model <- KMeansTrainer$new(clusters=2, batch_size=30, max_clusters=6) km_model$fit(data, find_optimal = FALSE) predictions <- km_model$predict(data)
Trains a k nearest neighbour model using fast search algorithms. KNN is a supervised learning algorithm which is used for both regression and classification problems.
R6Class object.
For usage details see Methods, Arguments and Examples sections.
bst = KNNTrainer$new(k=1, prob=FALSE, algorithm=NULL, type="class") bst$fit(X_train, X_test, "target") bst$predict(type)
$new()Initialise the instance of the trainer
$fit()trains the knn model and stores the test prediction
$predict()returns predictions
number of neighbours to predict
if probability should be computed, default=FALSE
algorithm used to train the model, possible values are 'kd_tree','cover_tree','brute'
type of problem to solve i.e. regression or classification, possible values are 'reg' or 'class'
knumber of neighbours to predict
probif probability should be computed, default=FALSE
algorithmalgorithm used to train the model, possible values are 'kd_tree','cover_tree','brute'
typetype of problem to solve i.e. regression or classification, possible values are 'reg' or 'class'
modelfor internal use
new()
KNNTrainer$new(k, prob, algorithm, type)
kk number of neighbours to predict
probif probability should be computed, default=FALSE
algorithmalgorithm used to train the model, possible values are 'kd_tree','cover_tree','brute'
typetype of problem to solve i.e. regression or classification, possible values are 'reg' or 'class'
Create a new 'KNNTrainer' object.
A 'KNNTrainer' object.
data("iris")
iris$Species <- as.integer(as.factor(iris$Species))
xtrain <- iris[1:100,]
xtest <- iris[101:150,]
bst <- KNNTrainer$new(k=3, prob=TRUE, type="class")
bst$fit(xtrain, xtest, 'Species')
pred <- bst$predict(type="raw")
fit()
KNNTrainer$fit(train, test, y)
traindata.frame or matrix
testdata.frame or matrix
ycharacter, name of target variable
Trains the KNNTrainer model
NULL
data("iris")
iris$Species <- as.integer(as.factor(iris$Species))
xtrain <- iris[1:100,]
xtest <- iris[101:150,]
bst <- KNNTrainer$new(k=3, prob=TRUE, type="class")
bst$fit(xtrain, xtest, 'Species')
predict()
KNNTrainer$predict(type = "raw")
typecharacter, 'raw' for labels else 'prob'
Predits the nearest neigbours for test data
a list of predicted neighbours
data("iris")
iris$Species <- as.integer(as.factor(iris$Species))
xtrain <- iris[1:100,]
xtest <- iris[101:150,]
bst <- KNNTrainer$new(k=3, prob=TRUE, type="class")
bst$fit(xtrain, xtest, 'Species')
pred <- bst$predict(type="raw")
clone()
The objects of this class are cloneable with this method.
KNNTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw") ## ------------------------------------------------ ## Method `KNNTrainer$new` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw") ## ------------------------------------------------ ## Method `KNNTrainer$fit` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') ## ------------------------------------------------ ## Method `KNNTrainer$predict` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw")data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw") ## ------------------------------------------------ ## Method `KNNTrainer$new` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw") ## ------------------------------------------------ ## Method `KNNTrainer$fit` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') ## ------------------------------------------------ ## Method `KNNTrainer$predict` ## ------------------------------------------------ data("iris") iris$Species <- as.integer(as.factor(iris$Species)) xtrain <- iris[1:100,] xtest <- iris[101:150,] bst <- KNNTrainer$new(k=3, prob=TRUE, type="class") bst$fit(xtrain, xtest, 'Species') pred <- bst$predict(type="raw")
Encodes and decodes categorical variables into integer values and vice versa. This is a commonly performed task in data preparation during model training, because all machine learning models require the data to be encoded into numerical format. It takes a vector of character or factor values and encodes them into numeric.
R6Class object.
For usage details see Methods, Arguments and Examples sections.
lbl = LabelEncoder$new() lbl$fit(x) lbl$fit_transform(x) lbl$transform(x)
$new()Initialise the instance of the encoder
$fit()creates a memory of encodings but doesn't return anything
$transform()based on encodings learned in fit method is applies the transformation
$fit_transform()encodes the data and keep a memory of encodings simultaneously
$inverse_transform()encodes the data and keep a memory of encodings simultaneously
a vector or list containing the character / factor values
input_datainternal use
encodingsinternal use
decodingsinternal use
fit_modelinternal use
fit()
LabelEncoder$fit(data_col)
data_cola vector containing non-null values
Fits the labelencoder model on given data
NULL, calculates the encoding and save in memory
data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co'))
lbl <- LabelEncoder$new()
lbl$fit(data_ex$Name)
data_ex$Name <- lbl$fit_transform(data_ex$Name)
decode_names <- lbl$inverse_transform(data_ex$Name)
fit_transform()
LabelEncoder$fit_transform(data_col)
data_cola vector containing non-null values
Fits and returns the encoding
encoding values for the given input data
data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co'))
lbl <- LabelEncoder$new()
lbl$fit(data_ex$Name)
data_ex$Name <- lbl$fit_transform(data_ex$Name)
transform()
LabelEncoder$transform(data_col)
data_cola vector containing non-null values
Returns the encodings from the fitted model
encoding values for the given input data
data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co'))
lbl <- LabelEncoder$new()
lbl$fit(data_ex$Name)
data_ex$Name <- lbl$transform(data_ex$Name)
inverse_transform()
LabelEncoder$inverse_transform(coded_col)
coded_cola vector containing label encoded values
Gives back the original values from a encoded values
original values from the label encoded data
data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co'))
lbl <- LabelEncoder$new()
lbl$fit(data_ex$Name)
data_ex$Name <- lbl$fit_transform(data_ex$Name)
decode_names <- lbl$inverse_transform(data_ex$Name)
clone()
The objects of this class are cloneable with this method.
LabelEncoder$clone(deep = FALSE)
deepWhether to make a deep clone.
data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$fit` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$fit_transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$inverse_transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name)data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$fit` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$fit_transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$transform(data_ex$Name) ## ------------------------------------------------ ## Method `LabelEncoder$inverse_transform` ## ------------------------------------------------ data_ex <- data.frame(Score = c(10,20,30,4), Name=c('Ao','Bo','Bo','Co')) lbl <- LabelEncoder$new() lbl$fit(data_ex$Name) data_ex$Name <- lbl$fit_transform(data_ex$Name) decode_names <- lbl$inverse_transform(data_ex$Name)
Trains regression, lasso, ridge model in R
Trains linear models such as Logistic, Lasso or Ridge regression model. It is built on glmnet R package. This class provides fit, predict, cross valdidation functions.
familytype of regression to perform, values can be "gaussian" ,"binomial", "multinomial","mgaussian"
weightsobservation weights. Can be total counts if responses are proportion matrices. Default is 1 for each observation
alphaThe elasticnet mixing parameter, alpha=1 is the lasso penalty, alpha=0 the ridge penalty, alpha=NULL is simple regression
lambdathe number of lambda values - default is 100
standardizenormalise the features in the given data
standardize.responsenormalise the dependent variable between 0 and 1, default = FALSE
modelinternal use
cvmodelinternal use
Flaginternal use
is_lassointernal use
iid_namesinternal use
new()
LMTrainer$new(family, weights, alpha, lambda, standardize.response)
familycharacter, type of regression to perform, values can be "gaussian" ,"binomial", "multinomial","mgaussian"
weightsnumeric, observation weights. Can be total counts if responses are proportion matrices. Default is 1 for each observation
alphainteger, The elasticnet mixing parameter, alpha=1 is the lasso penalty, alpha=0 the ridge penalty, alpha=NULL is simple regression
lambdainteger, the number of lambda values - default is 100
standardize.responselogical, normalise the dependent variable between 0 and 1, default = FALSE
Create a new 'LMTrainer' object.
A 'LMTrainer' object.
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
}
fit()
LMTrainer$fit(X, y)
Xdata.frame containing train featuers
ycharacter, name of target variable
Fits the LMTrainer model on given data
NULL, train the model and saves internally
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
lf$fit(X = housing, y = 'MEDV')
}
predict()
LMTrainer$predict(df, lambda = NULL)
dfdata.frame containing test features
lambdainteger, the number of lambda values - default is 100. By default it picks the best value from the model.
Returns predictions for test data
vector, a vector containing predictions
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
lf$fit(X = housing, y = 'MEDV')
predictions <- lf$cv_predict(df = housing)
}
cv_model()
LMTrainer$cv_model(X, y, nfolds, parallel, type.measure = "deviance")
Xdata.frame containing test features
ycharacter, name of target variable
nfoldsinteger, number of folds
parallellogical, if do parallel computation. Default=FALSE
type.measurecharacter, evaluation metric type. Default = deviance
Train regression model using cross validation
NULL, trains the model and saves it in memory
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE)
}
cv_predict()
LMTrainer$cv_predict(df, lambda = NULL)
dfdata.frame containing test features
lambdainteger, the number of lambda values - default is 100. By default it picks the best value from the model.
Get predictions from the cross validated regression model
vector a vector containing predicted values
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE)
predictions <- lf$cv_predict(df = housing)
}
get_importance()
LMTrainer$get_importance()
Get feature importance using model coefficients
a matrix containing feature coefficients
\dontrun{
LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data"
housing <- read.table(LINK)
names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS",
"RAD","TAX","PTRATIO","B","LSTAT","MEDV")
names(housing) <- names
lf <- LMTrainer$new(family = 'gaussian', alpha=1)
lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE)
predictions <- lf$cv_predict(df = housing)
coefs <- lf$get_importance()
}
clone()
The objects of this class are cloneable with this method.
LMTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `LMTrainer$new` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$fit` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$fit(X = housing, y = 'MEDV') ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$predict` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$fit(X = housing, y = 'MEDV') predictions <- lf$cv_predict(df = housing) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$cv_model` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$cv_predict` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) predictions <- lf$cv_predict(df = housing) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$get_importance` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) predictions <- lf$cv_predict(df = housing) coefs <- lf$get_importance() ## End(Not run)## ------------------------------------------------ ## Method `LMTrainer$new` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$fit` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$fit(X = housing, y = 'MEDV') ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$predict` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$fit(X = housing, y = 'MEDV') predictions <- lf$cv_predict(df = housing) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$cv_model` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$cv_predict` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) predictions <- lf$cv_predict(df = housing) ## End(Not run) ## ------------------------------------------------ ## Method `LMTrainer$get_importance` ## ------------------------------------------------ ## Not run: LINK <- "http://archive.ics.uci.edu/ml/machine-learning-databases/housing/housing.data" housing <- read.table(LINK) names <- c("CRIM","ZN","INDUS","CHAS","NOX","RM","AGE","DIS", "RAD","TAX","PTRATIO","B","LSTAT","MEDV") names(housing) <- names lf <- LMTrainer$new(family = 'gaussian', alpha=1) lf$cv_model(X = housing, y = 'MEDV', nfolds = 5, parallel = FALSE) predictions <- lf$cv_predict(df = housing) coefs <- lf$get_importance() ## End(Not run)
Trains a probabilistic naive bayes model
Trains a naive bayes model. It is built on top high performance naivebayes R package.
priornumeric vector with prior probabilities. vector with prior probabilities of the classes. If unspecified, the class proportions for the training set are used. If present, the probabilities should be specified in the order of the factor levels.
laplacevalue used for Laplace smoothing. Defaults to 0 (no Laplace smoothing)
usekernelif TRUE, density is used to estimate the densities of metric predictors
modelfor internal use
new()
NBTrainer$new(prior, laplace, usekernel)
priornumeric, prior numeric vector with prior probabilities. vector with prior probabilities of the classes. If unspecified, the class proportions for the training set are used. If present, the probabilities should be specified in the order of the factor levels.
laplacenuemric, value used for Laplace smoothing. Defaults to 0 (no Laplace smoothing)
usekernellogical, if TRUE, density is used to estimate the densities of metric predictors
Create a new 'NBTrainer' object.
A 'NBTrainer' object.
data(iris) nb <- NBTrainer$new()
fit()
NBTrainer$fit(X, y)
Xdata.frame containing train features
ycharacter, name of target variable
Fits the naive bayes model
NULL, trains and saves the model in memory
data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species')
predict()
NBTrainer$predict(X, type = "class")
Xdata.frame containing test features
typecharacter, if the predictions should be labels or probability
Returns predictions from the model
NULL, trains and saves the model in memory
data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species') y <- nb$predict(iris)
clone()
The objects of this class are cloneable with this method.
NBTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `NBTrainer$new` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() ## ------------------------------------------------ ## Method `NBTrainer$fit` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species') ## ------------------------------------------------ ## Method `NBTrainer$predict` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species') y <- nb$predict(iris)## ------------------------------------------------ ## Method `NBTrainer$new` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() ## ------------------------------------------------ ## Method `NBTrainer$fit` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species') ## ------------------------------------------------ ## Method `NBTrainer$predict` ## ------------------------------------------------ data(iris) nb <- NBTrainer$new() nb$fit(iris, 'Species') y <- nb$predict(iris)
Normalises a 1 dimensional vector towards unit p norm. By default, p = 2 is used. For a given vector, eg: c(1,2,3), norm value is calculated as 'x / |x|' where '|x|' is calculated as the square root of sum of square of values in the given vector.
normalise1d(vec, pnorm = 2L)normalise1d(vec, pnorm = 2L)
vec |
vector containing integers or numeric values. |
pnorm |
integer, default: 2 |
a vector containing normalised values
val <- c(1,10,5,3,8) norm_val <- normalise1d(val)val <- c(1,10,5,3,8) norm_val <- normalise1d(val)
Normalises a matrix towards unit p norm row wise or column wise. By default, p = 2 is used. To normalise row wise, use axis=0. To normalise column wise, use axis=1. as the square root of sum of square of values in the given vector.
normalise2d(mat, pnorm = 2L, axis = 1L)normalise2d(mat, pnorm = 2L, axis = 1L)
mat |
numeric matrix |
pnorm |
integer value, default value=2 |
axis |
integer (0 or 1), row wise = 0, column wise = 1 |
normalised numeric matrix
mat <- matrix(runif(12), 3, 4) ## normalise matrix row wise r <- normalise2d(mat, axis=0) ## normalise matrix column wise r <- normalise2d(mat, axis=1)mat <- matrix(runif(12), 3, 4) ## normalise matrix row wise r <- normalise2d(mat, axis=0) ## normalise matrix column wise r <- normalise2d(mat, axis=1)
Hyperparameter tuning using random search scheme.
Given a set of hyper parameters, random search trainer provides a faster way of hyper parameter tuning. Here, the number of models to be trained can be defined by the user.
superml::GridSearchCV -> RandomSearchTrainer
n_iternumber of models to be trained
new()
RandomSearchCV$new( trainer = NA, parameters = NA, n_folds = NA, scoring = NA, n_iter = NA )
trainersuperml trainer object, must be either XGBTrainer, LMTrainer, RFTrainer, NBTrainer
parameterslist, list containing parameters
n_foldsinteger, number of folds to use to split the train data
scoringcharacter, scoring metric used to evaluate the best model, multiple values can be provided. currently supports: auc, accuracy, mse, rmse, logloss, mae, f1, precision, recall
n_iterinteger, number of models to be trained
Create a new 'RandomSearchTrainer' object.
A 'RandomSearchTrainer' object.
rf <- RFTrainer$new()
rst <-RandomSearchCV$new(trainer = rf,
parameters = list(n_estimators = c(100,500),
max_depth = c(5,2,10,14)),
n_folds = 3,
scoring = c('accuracy','auc'),
n_iter = 4)
fit()
RandomSearchCV$fit(X, y)
Xdata.frame containing features
ycharacter, name of target variable
Train the model on given hyperparameters
NULL, tunes hyperparameters and stores the result in memory
rf <- RFTrainer$new()
rst <-RandomSearchCV$new(trainer = rf,
parameters = list(n_estimators = c(100,500),
max_depth = c(5,2,10,14)),
n_folds = 3,
scoring = c('accuracy','auc'),
n_iter = 4)
data("iris")
rst$fit(iris, "Species")
rst$best_iteration()
clone()
The objects of this class are cloneable with this method.
RandomSearchCV$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `RandomSearchCV$new` ## ------------------------------------------------ rf <- RFTrainer$new() rst <-RandomSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100,500), max_depth = c(5,2,10,14)), n_folds = 3, scoring = c('accuracy','auc'), n_iter = 4) ## ------------------------------------------------ ## Method `RandomSearchCV$fit` ## ------------------------------------------------ rf <- RFTrainer$new() rst <-RandomSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100,500), max_depth = c(5,2,10,14)), n_folds = 3, scoring = c('accuracy','auc'), n_iter = 4) data("iris") rst$fit(iris, "Species") rst$best_iteration()## ------------------------------------------------ ## Method `RandomSearchCV$new` ## ------------------------------------------------ rf <- RFTrainer$new() rst <-RandomSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100,500), max_depth = c(5,2,10,14)), n_folds = 3, scoring = c('accuracy','auc'), n_iter = 4) ## ------------------------------------------------ ## Method `RandomSearchCV$fit` ## ------------------------------------------------ rf <- RFTrainer$new() rst <-RandomSearchCV$new(trainer = rf, parameters = list(n_estimators = c(100,500), max_depth = c(5,2,10,14)), n_folds = 3, scoring = c('accuracy','auc'), n_iter = 4) data("iris") rst$fit(iris, "Species") rst$best_iteration()
Training Dataset used for regression examples. In this data set, we have to predict the sale price of the houses.
reg_trainreg_train
An object of class data.table (inherits from data.frame) with 1460 rows and 81 columns.
https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data
Trains a random forest model.
Trains a Random Forest model. A random forest is a meta estimator that fits a number of decision tree classifiers on various sub-samples of the dataset and use averaging to improve the predictive accuracy and control over-fitting. This implementation uses ranger R package which provides faster model training.
n_estimatorsthe number of trees in the forest, default= 100
max_featuresthe number of features to consider when looking for the best split.
Possible values are auto(default) takes sqrt(num_of_features),
sqrt same as auto,
log takes log(num_of_features),
none takes all features
max_depththe maximum depth of each tree
min_node_sizethe minumum number of samples required to split an internal node
criterionthe function to measure the quality of split. For classification, gini is used which
is a measure of gini index. For regression, the variance of responses is used.
classificationwhether to train for classification (1) or regression (0)
verboseshow computation status and estimated runtime
seedseed value
class_weightsweights associated with the classes for sampling of training observation
always_splitvector of feature names to be always used for splitting
importanceVariable importance mode, one of 'none', 'impurity', 'impurity_corrected', 'permutation'. The 'impurity' measure is the Gini index for classification, the variance of the responses for regression. Defaults to "impurity"
new()
RFTrainer$new( n_estimators, max_depth, max_features, min_node_size, classification, class_weights, always_split, verbose, save_model, seed, importance )
n_estimatorsinteger, the number of trees in the forest, default= 100
max_depthinteger, the maximum depth of each tree
max_featuresinteger, the number of features to consider when looking for the best split.
Possible values are auto(default) takes sqrt(num_of_features),
sqrt same as auto,
log takes log(num_of_features),
none takes all features
min_node_sizeinteger, the minumum number of samples required to split an internal node
classificationinteger, whether to train for classification (1) or regression (0)
class_weightsweights associated with the classes for sampling of training observation
always_splitvector of feature names to be always used for splitting
verboselogical, show computation status and estimated runtime
save_modellogical, whether to save model
seedinteger, seed value
importanceVariable importance mode, one of 'none', 'impurity', 'impurity_corrected', 'permutation'. The 'impurity' measure is the Gini index for classification, the variance of the responses for regression. Defaults to "impurity"
Create a new 'RFTrainer' object.
A 'RFTrainer' object.
data("iris")
bst <- RFTrainer$new(n_estimators=10,
max_depth=4,
classification=1,
seed=42,
verbose=TRUE)
fit()
RFTrainer$fit(X, y)
Xdata.frame containing train features
ycharacter, name of the target variable
Trains the random forest model
NULL, trains and saves the model in memory
data("iris")
bst <- RFTrainer$new(n_estimators=10,
max_depth=4,
classification=1,
seed=42,
verbose=TRUE)
bst$fit(iris, 'Species')
predict()
RFTrainer$predict(df)
dfdata.frame containing test features
Return predictions from random forest model
a vector containing predictions
data("iris")
bst <- RFTrainer$new(n_estimators=10,
max_depth=4,
classification=1,
seed=42,
verbose=TRUE)
bst$fit(iris, 'Species')
predictions <- bst$predict(iris)
get_importance()
RFTrainer$get_importance()
Returns feature importance from the model
a data frame containing feature predictions
data("iris")
bst <- RFTrainer$new(n_estimators=50,
max_depth=4,
classification=1,
seed=42,
verbose=TRUE)
bst$fit(iris, 'Species')
predictions <- bst$predict(iris)
bst$get_importance()
clone()
The objects of this class are cloneable with this method.
RFTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `RFTrainer$new` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) ## ------------------------------------------------ ## Method `RFTrainer$fit` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') ## ------------------------------------------------ ## Method `RFTrainer$predict` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') predictions <- bst$predict(iris) ## ------------------------------------------------ ## Method `RFTrainer$get_importance` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=50, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') predictions <- bst$predict(iris) bst$get_importance()## ------------------------------------------------ ## Method `RFTrainer$new` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) ## ------------------------------------------------ ## Method `RFTrainer$fit` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') ## ------------------------------------------------ ## Method `RFTrainer$predict` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=10, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') predictions <- bst$predict(iris) ## ------------------------------------------------ ## Method `RFTrainer$get_importance` ## ------------------------------------------------ data("iris") bst <- RFTrainer$new(n_estimators=50, max_depth=4, classification=1, seed=42, verbose=TRUE) bst$fit(iris, 'Species') predictions <- bst$predict(iris) bst$get_importance()
Calculates target encodings using a smoothing parameter and count of categorical variables. This approach is more robust to possibility of leakage and avoid overfitting.
smoothMean( train_df, test_df, colname, target, min_samples_leaf = 1, smoothing = 1, noise_level = 0 )smoothMean( train_df, test_df, colname, target, min_samples_leaf = 1, smoothing = 1, noise_level = 0 )
train_df |
train dataset |
test_df |
test dataset |
colname |
name of categorical column |
target |
name of target column |
min_samples_leaf |
minimum samples to take category average into account |
smoothing |
smoothing effect to balance categorical average vs prior |
noise_level |
random noise to add, optional |
a train and test data table with mean encodings of the target for the given categorical variable
train <- data.frame(region=c('del','csk','rcb','del','csk','pune','guj','del'), win = c(0,1,1,0,0,1,0,1)) test <- data.frame(region=c('rcb','csk','rcb','del','guj','pune','csk','kol')) # calculate encodings all_means <- smoothMean(train_df = train, test_df = test, colname = 'region', target = 'win') train_mean <- all_means$train test_mean <- all_means$testtrain <- data.frame(region=c('del','csk','rcb','del','csk','pune','guj','del'), win = c(0,1,1,0,0,1,0,1)) test <- data.frame(region=c('rcb','csk','rcb','del','guj','pune','csk','kol')) # calculate encodings all_means <- smoothMean(train_df = train, test_df = test, colname = 'region', target = 'win') train_mean <- all_means$train test_mean <- all_means$test
For a given vector, return the indexes of the sorted array and not the sorted array itself.
sort_index(vec, ascending = TRUE)sort_index(vec, ascending = TRUE)
vec |
numeric vector |
ascending |
logical, order to return (ascending or descending), default = True |
numeric vector containing sorted indexes
v <- c(10,3,1,4) j <- sort_index(v)v <- c(10,3,1,4) j <- sort_index(v)
Creates a tf-idf matrix
Given a list of text, it creates a sparse matrix consisting of tf-idf score for tokens from the text.
superml::CountVectorizer -> TfIdfVectorizer
sentencesa list containing sentences
max_dfWhen building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold, value lies between 0 and 1.
min_dfWhen building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold, value lies between 0 and 1.
max_featuresuse top features sorted by count to be used in bag of words matrix.
ngram_rangeThe lower and upper boundary of the range of n-values for different word n-grams or char n-grams to be extracted. All values of n such such that min_n <= n <= max_n will be used. For example an ngram_range of c(1, 1) means only unigrams, c(1, 2) means unigrams and bigrams, and c(2, 2) means only bigrams.
splitsplitting criteria for strings, default: " "
lowercaseconvert all characters to lowercase before tokenizing
regexregex expression to use for text cleaning.
remove_stopwordsa list of stopwords to use, by default it uses its inbuilt list of standard stopwords
smooth_idflogical, to prevent zero division, adds one to document frequencies, as if an extra document was seen containing every term in the collection exactly once
normlogical, if TRUE, each output row will have unit norm ‘l2’: Sum of squares of vector elements is 1. if FALSE returns non-normalized vectors, default: TRUE
new()
TfIdfVectorizer$new( min_df, max_df, max_features, ngram_range, regex, remove_stopwords, split, lowercase, smooth_idf, norm )
min_dfnumeric, When building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold, value lies between 0 and 1.
max_dfnumeric, When building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold, value lies between 0 and 1.
max_featuresinteger, Build a vocabulary that only consider the top max_features ordered by term frequency across the corpus.
ngram_rangevector, The lower and upper boundary of the range of n-values for different word n-grams or char n-grams to be extracted. All values of n such such that min_n <= n <= max_n will be used. For example an ngram_range of c(1, 1) means only unigrams, c(1, 2) means unigrams and bigrams, and c(2, 2) means only bigrams.
regexcharacter, regex expression to use for text cleaning.
remove_stopwordslist, a list of stopwords to use, by default it uses its inbuilt list of standard english stopwords
splitcharacter, splitting criteria for strings, default: " "
lowercaselogical, convert all characters to lowercase before tokenizing, default: TRUE
smooth_idflogical, to prevent zero division, adds one to document frequencies, as if an extra document was seen containing every term in the collection exactly once
normlogical, if TRUE, each output row will have unit norm ‘l2’: Sum of squares of vector elements is 1. if FALSE returns non-normalized vectors, default: TRUE
parallellogical, speeds up ngrams computation using n-1 cores, defaults: TRUE
Create a new 'TfIdfVectorizer' object.
A 'TfIdfVectorizer' object.
TfIdfVectorizer$new()
fit()
TfIdfVectorizer$fit(sentences)
sentencesa list of text sentences
Fits the TfIdfVectorizer model on sentences
NULL
sents = c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
tf = TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.3)
tf$fit(sents)
fit_transform()
TfIdfVectorizer$fit_transform(sentences)
sentencesa list of text sentences
Fits the TfIdfVectorizer model and returns a sparse matrix of count of tokens
a sparse matrix containing tf-idf score for tokens in each given sentence
\dontrun{
sents <- c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
tf <- TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.1)
tf_matrix <- tf$fit_transform(sents)
}
transform()
TfIdfVectorizer$transform(sentences)
sentencesa list of new text sentences
Returns a matrix of tf-idf score of tokens
a sparse matrix containing tf-idf score for tokens in each given sentence
\dontrun{
sents = c('i am alone in dark.','mother_mary a lot',
'alone in the dark?', 'many mothers in the lot....')
new_sents <- c("dark at night",'mothers day')
tf = TfIdfVectorizer$new(min_df=0.1)
tf$fit(sents)
tf_matrix <- tf$transform(new_sents)
}
clone()
The objects of this class are cloneable with this method.
TfIdfVectorizer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `TfIdfVectorizer$new` ## ------------------------------------------------ TfIdfVectorizer$new() ## ------------------------------------------------ ## Method `TfIdfVectorizer$fit` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') tf = TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.3) tf$fit(sents) ## ------------------------------------------------ ## Method `TfIdfVectorizer$fit_transform` ## ------------------------------------------------ ## Not run: sents <- c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') tf <- TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.1) tf_matrix <- tf$fit_transform(sents) ## End(Not run) ## ------------------------------------------------ ## Method `TfIdfVectorizer$transform` ## ------------------------------------------------ ## Not run: sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') new_sents <- c("dark at night",'mothers day') tf = TfIdfVectorizer$new(min_df=0.1) tf$fit(sents) tf_matrix <- tf$transform(new_sents) ## End(Not run)## ------------------------------------------------ ## Method `TfIdfVectorizer$new` ## ------------------------------------------------ TfIdfVectorizer$new() ## ------------------------------------------------ ## Method `TfIdfVectorizer$fit` ## ------------------------------------------------ sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') tf = TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.3) tf$fit(sents) ## ------------------------------------------------ ## Method `TfIdfVectorizer$fit_transform` ## ------------------------------------------------ ## Not run: sents <- c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') tf <- TfIdfVectorizer$new(smooth_idf = TRUE, min_df = 0.1) tf_matrix <- tf$fit_transform(sents) ## End(Not run) ## ------------------------------------------------ ## Method `TfIdfVectorizer$transform` ## ------------------------------------------------ ## Not run: sents = c('i am alone in dark.','mother_mary a lot', 'alone in the dark?', 'many mothers in the lot....') new_sents <- c("dark at night",'mothers day') tf = TfIdfVectorizer$new(min_df=0.1) tf$fit(sents) tf_matrix <- tf$transform(new_sents) ## End(Not run)
Trains a XGBoost model in R
Trains a Extreme Gradient Boosting Model. XGBoost belongs to a family of boosting algorithms that creates an ensemble of weak learner to learn about data. It is a wrapper for original xgboost R package, you can find the documentation here: http://xgboost.readthedocs.io/en/latest/parameter.html
boosterthe trainer type, the values are gbtree(default), gblinear, dart:gbtree
objectivespecify the learning task. Check the link above for all possible values.
nthreadnumber of parallel threads used to run, default is to run using all threads available
silent0 means printing running messages, 1 means silent mode
n_estimatorsnumber of trees to grow, default = 100
learning_rateStep size shrinkage used in update to prevents overfitting. Lower the learning rate, more time it takes in training, value lies between between 0 and 1. Default = 0.3
gammaMinimum loss reduction required to make a further partition on a leaf node of the tree. The larger gamma is, the more conservative the algorithm will be. Value lies between 0 and infinity, Default = 0
max_depththe maximum depth of each tree, default = 6
min_child_weightMinimum sum of instance weight (hessian) needed in a child. If the tree partition step results in a leaf node with the sum of instance weight less than min_child_weight, then the building process will give up further partitioning. In linear regression task, this simply corresponds to minimum number of instances needed to be in each node. The larger min_child_weight is, the more conservative the algorithm will be. Value lies between 0 and infinity. Default = 1
subsampleSubsample ratio of the training instances. Setting it to 0.5 means that XGBoost would randomly sample half of the training data prior to growing trees. and this will prevent overfitting. Subsampling will occur once in every boosting iteration. Value lies between 0 and 1. Default = 1
colsample_bytreeSubsample ratio of columns when constructing each tree. Subsampling will occur once in every boosting iteration. Value lies between 0 and 1. Default = 1
lambdaL2 regularization term on weights. Increasing this value will make model more conservative. Default = 1
alphaL1 regularization term on weights. Increasing this value will make model more conservative. Default = 0
eval_metricEvaluation metrics for validation data, a default metric will be assigned according to objective
print_everyprint training log after n iterations. Default = 50
fevalcustom evaluation function
early_stoppingUsed to prevent overfitting, stops model training after this number of iterations if there is no improvement seen
maximizeIf feval and early_stopping_rounds are set, then this parameter must be set as well. When it is TRUE, it means the larger the evaluation score the better.
custom_objectivecustom objective function
save_periodwhen it is non-NULL, model is saved to disk after every save_period rounds, 0 means save at the end.
save_namethe name or path for periodically saved model file.
xgb_modela previously built model to continue the training from. Could be either an object of class xgb.Booster, or its raw data, or the name of a file with a previously saved model.
callbacksa list of callback functions to perform various task during boosting. See callbacks. Some of the callbacks are automatically created depending on the parameters' values. User can provide either existing or their own callback methods in order to customize the training process.
verboseIf 0, xgboost will stay silent. If 1, xgboost will print information of performance. If 2, xgboost will print some additional information. Setting verbose > 0 automatically engages the cb.evaluation.log and cb.print.evaluation callback functions.
watchlistwhat information should be printed when verbose=1 or verbose=2. Watchlist is used to specify validation set monitoring during training. For example user can specify watchlist=list(validation1=mat1, validation2=mat2) to watch the performance of each round's model on mat1 and mat2
num_classset number of classes in case of multiclassification problem
weighta vector indicating the weight for each row of the input.
na_missingby default is set to NA, which means that NA values should be considered as 'missing' by the algorithm. Sometimes, 0 or other extreme value might be used to represent missing values. This parameter is only used when input is a dense matrix.
feature_namesinternal use, stores the feature names for model importance
cv_modelinternal use
new()
XGBTrainer$new( booster, objective, nthread, silent, n_estimators, learning_rate, gamma, max_depth, min_child_weight, subsample, colsample_bytree, lambda, alpha, eval_metric, print_every, feval, early_stopping, maximize, custom_objective, save_period, save_name, xgb_model, callbacks, verbose, num_class, weight, na_missing )
boosterthe trainer type, the values are gbtree(default), gblinear, dart:gbtree
objectivespecify the learning task. Check the link above for all possible values.
nthreadnumber of parallel threads used to run, default is to run using all threads available
silent0 means printing running messages, 1 means silent mode
n_estimatorsnumber of trees to grow, default = 100
learning_rateStep size shrinkage used in update to prevents overfitting. Lower the learning rate, more time it takes in training, value lies between between 0 and 1. Default = 0.3
gammaMinimum loss reduction required to make a further partition on a leaf node of the tree. The larger gamma is, the more conservative the algorithm will be. Value lies between 0 and infinity, Default = 0
max_depththe maximum depth of each tree, default = 6
min_child_weightMinimum sum of instance weight (hessian) needed in a child. If the tree partition step results in a leaf node with the sum of instance weight less than min_child_weight, then the building process will give up further partitioning. In linear regression task, this simply corresponds to minimum number of instances needed to be in each node. The larger min_child_weight is, the more conservative the algorithm will be. Value lies between 0 and infinity. Default = 1
subsampleSubsample ratio of the training instances. Setting it to 0.5 means that XGBoost would randomly sample half of the training data prior to growing trees. and this will prevent overfitting. Subsampling will occur once in every boosting iteration. Value lies between 0 and 1. Default = 1
colsample_bytreeSubsample ratio of columns when constructing each tree. Subsampling will occur once in every boosting iteration. Value lies between 0 and 1. Default = 1
lambdaL2 regularization term on weights. Increasing this value will make model more conservative. Default = 1
alphaL1 regularization term on weights. Increasing this value will make model more conservative. Default = 0
eval_metricEvaluation metrics for validation data, a default metric will be assigned according to objective
print_everyprint training log after n iterations. Default = 50
fevalcustom evaluation function
early_stoppingUsed to prevent overfitting, stops model training after this number of iterations if there is no improvement seen
maximizeIf feval and early_stopping_rounds are set, then this parameter must be set as well. When it is TRUE, it means the larger the evaluation score the better.
custom_objectivecustom objective function
save_periodwhen it is non-NULL, model is saved to disk after every save_period rounds, 0 means save at the end.
save_namethe name or path for periodically saved model file.
xgb_modela previously built model to continue the training from. Could be either an object of class xgb.Booster, or its raw data, or the name of a file with a previously saved model.
callbacksa list of callback functions to perform various task during boosting. See callbacks. Some of the callbacks are automatically created depending on the parameters' values. User can provide either existing or their own callback methods in order to customize the training process.
verboseIf 0, xgboost will stay silent. If 1, xgboost will print information of performance. If 2, xgboost will print some additional information. Setting verbose > 0 automatically engages the cb.evaluation.log and cb.print.evaluation callback functions.
num_classset number of classes in case of multiclassification problem
weighta vector indicating the weight for each row of the input.
na_missingby default is set to NA, which means that NA values should be considered as 'missing' by the algorithm. Sometimes, 0 or other extreme value might be used to represent missing values. This parameter is only used when input is a dense matrix.
Create a new 'XGBTrainer' object.
A 'XGBTrainer' object.
library(data.table)
df <- copy(iris)
# convert characters/factors to numeric
df$Species <- as.numeric(as.factor(df$Species))-1
# initialise model
xgb <- XGBTrainer$new(objective = 'multi:softmax',
maximize = FALSE,
eval_metric = 'merror',
num_class=3,
n_estimators = 2)
cross_val()
XGBTrainer$cross_val(X, y, nfolds = 5, stratified = TRUE, folds = NULL)
Xdata.frame
ycharacter, name of target variable
nfoldsinteger, number of folds
stratifiedlogical, whether to use stratified sampling
foldsthe list of CV folds' indices - either those passed through the folds parameter or randomly generated.
Trains the xgboost model using cross validation scheme
NULL, trains a model and saves it in memory
\dontrun{
library(data.table)
df <- copy(iris)
# convert characters/factors to numeric
df$Species <- as.numeric(as.factor(df$Species))-1
# initialise model
xgb <- XGBTrainer$new(objective = 'multi:softmax',
maximize = FALSE,
eval_metric = 'merror',
num_class=3,
n_estimators = 2)
# do cross validation to find optimal value for n_estimators
xgb$cross_val(X = df, y = 'Species',nfolds = 3, stratified = TRUE)
}
fit()
XGBTrainer$fit(X, y, valid = NULL)
Xdata.frame, training data
ycharacter, name of target variable
validdata.frame, validation data
Fits the xgboost model on given data
NULL, trains a model and keeps it in memory
library(data.table)
df <- copy(iris)
# convert characters/factors to numeric
df$Species <- as.numeric(as.factor(df$Species))-1
# initialise model
xgb <- XGBTrainer$new(objective = 'multi:softmax',
maximize = FALSE,
eval_metric = 'merror',
num_class=3,
n_estimators = 2)
xgb$fit(df, 'Species')
predict()
XGBTrainer$predict(df)
dfdata.frame, test data set
Returns predicted values for a given test data
xgboost predictions
#' library(data.table)
df <- copy(iris)
# convert characters/factors to numeric
df$Species <- as.numeric(as.factor(df$Species))-1
# initialise model
xgb <- XGBTrainer$new(objective = 'multi:softmax',
maximize = FALSE,
eval_metric = 'merror',
num_class=3,
n_estimators = 2)
xgb$fit(df, 'Species')
# make predictions
preds <- xgb$predict(as.matrix(iris[,1:4]))
show_importance()
XGBTrainer$show_importance(type = "plot", topn = 10)
typecharacter, could be 'plot' or 'table'
topninteger, top n features to display
Shows feature importance plot
a table or a plot of feature importance
\dontrun{
library(data.table)
df <- copy(iris)
# convert characters/factors to numeric
df$Species <- as.numeric(as.factor(df$Species))-1
# initialise model
xgb <- XGBTrainer$new(objective = 'multi:softmax',
maximize = FALSE,
eval_metric = 'merror',
num_class=3,
n_estimators = 2)
xgb$fit(df, 'Species')
xgb$show_importance()
}
clone()
The objects of this class are cloneable with this method.
XGBTrainer$clone(deep = FALSE)
deepWhether to make a deep clone.
## ------------------------------------------------ ## Method `XGBTrainer$new` ## ------------------------------------------------ library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) ## ------------------------------------------------ ## Method `XGBTrainer$cross_val` ## ------------------------------------------------ ## Not run: library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) # do cross validation to find optimal value for n_estimators xgb$cross_val(X = df, y = 'Species',nfolds = 3, stratified = TRUE) ## End(Not run) ## ------------------------------------------------ ## Method `XGBTrainer$fit` ## ------------------------------------------------ library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') ## ------------------------------------------------ ## Method `XGBTrainer$predict` ## ------------------------------------------------ #' library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') # make predictions preds <- xgb$predict(as.matrix(iris[,1:4])) ## ------------------------------------------------ ## Method `XGBTrainer$show_importance` ## ------------------------------------------------ ## Not run: library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') xgb$show_importance() ## End(Not run)## ------------------------------------------------ ## Method `XGBTrainer$new` ## ------------------------------------------------ library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) ## ------------------------------------------------ ## Method `XGBTrainer$cross_val` ## ------------------------------------------------ ## Not run: library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) # do cross validation to find optimal value for n_estimators xgb$cross_val(X = df, y = 'Species',nfolds = 3, stratified = TRUE) ## End(Not run) ## ------------------------------------------------ ## Method `XGBTrainer$fit` ## ------------------------------------------------ library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') ## ------------------------------------------------ ## Method `XGBTrainer$predict` ## ------------------------------------------------ #' library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') # make predictions preds <- xgb$predict(as.matrix(iris[,1:4])) ## ------------------------------------------------ ## Method `XGBTrainer$show_importance` ## ------------------------------------------------ ## Not run: library(data.table) df <- copy(iris) # convert characters/factors to numeric df$Species <- as.numeric(as.factor(df$Species))-1 # initialise model xgb <- XGBTrainer$new(objective = 'multi:softmax', maximize = FALSE, eval_metric = 'merror', num_class=3, n_estimators = 2) xgb$fit(df, 'Species') xgb$show_importance() ## End(Not run)