model_text_regressor {autokeras}

AutoKeras Text Regressor Model

Description

AutoKeras text regression class.
To 'fit', 'evaluate' or 'predict', format inputs as:

• x : array. The input data should be array. The data should be one dimensional. Each element in the data should be a string which is a full sentence.

• y : array. The targets passing to the head would have to be array or data.frame. It can be single-column or multi-column. The values should all be numerical.

Usage

model_text_regressor(
  output_dim = NULL,
  loss = "mean_squared_error",
  metrics = NULL,
  name = "text_regressor",
  max_trials = 100,
  directory = tempdir(),
  objective = "val_loss",
  overwrite = TRUE,
  seed = runif(1, 0, 1e+07)
)

Arguments

output_dim

: numeric. The number of output dimensions. Defaults to 'NULL'. If 'NULL', it will infer from the data.

loss

: A Keras loss function. Defaults to use "mean_squared_error".

metrics

: A list of Keras metrics. Defaults to use "mean_squared_error".

name

: character. The name of the AutoModel. Defaults to "text_regressor".

max_trials

: numeric. The maximum number of different Keras Models to try. The search may finish before reaching the 'max_trials'. Defaults to '100'.

directory

: character. The path to a directory for storing the search outputs. Defaults to 'tempdir()', which would create a folder with the name of the AutoModel in the current directory.

objective

: character. Name of model metric to minimize or maximize, e.g. "val_accuracy". Defaults to "val_loss".

overwrite

: logical. Defaults to 'TRUE'. If 'FALSE', reloads an existing project of the same name if one is found. Otherwise, overwrites the project.

seed

: numeric. Random seed. Defaults to 'runif(1, 0, 10e6)'.

Details

Important: The object returned by this function behaves like an R6 object, i.e., within function calls with this object as parameter, it is most likely that the object will be modified. Therefore it is not necessary to assign the result of the functions to the same object.

Value

A non-trained text regressor AutokerasModel.

Examples

## Not run: 
library("keras")

# Get IMDb dataset
imdb <- dataset_imdb(num_words = 1000)
c(x_train, y_train) %<-% imdb$train
c(x_test, y_test) %<-% imdb$test

# AutoKeras procceses each text data point as a character vector,
# i.e., x_train[[1]] "<START> this film was just brilliant casting..",
# so we need to transform the dataset.
word_index <- dataset_imdb_word_index()
word_index <- c(
  "<PAD>", "<START>", "<UNK>", "<UNUSED>",
  names(word_index)[order(unlist(word_index))]
)
x_train <- lapply(x_train, function(x) {
  paste(word_index[x + 1], collapse = " ")
})
x_test <- lapply(x_test, function(x) {
  paste(word_index[x + 1], collapse = " ")
})

x_train <- matrix(unlist(x_train), ncol = 1)
x_test <- matrix(unlist(x_test), ncol = 1)
y_train <- array(unlist(y_train))
y_test <- array(unlist(y_test))

library("autokeras")

# Initialize the text regressor
reg <- model_text_regressor(max_trials = 10) %>% # It tries 10 different models
  fit(x_train, y_train) # Feed the text regressor with training data

# If you want to use own valitadion data do:
reg <- model_text_regressor(max_trials = 10) %>%
  fit(
    x_train,
    y_train,
    validation_data = list(x_test, y_test)
  )

# Predict with the best model
(predicted_y <- reg %>% predict(x_test))

# Evaluate the best model with testing data
reg %>% evaluate(x_test, y_test)

# Get the best trained Keras model, to work with the keras R library
export_model(reg)

## End(Not run)