Skip to content

Text Classification

View in Colab     GitHub source

!pip install autokeras
!pip install git+https://github.com/keras-team/keras-tuner.git@1.0.2rc3

A Simple Example

The first step is to prepare your data. Here we use the IMDB dataset as an example.

import os
import numpy as np
import tensorflow as tf
from tensorflow.keras.datasets import imdb
from sklearn.datasets import load_files

dataset = tf.keras.utils.get_file(
    fname="aclImdb.tar.gz",
    origin="http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz",
    extract=True,
)

# set path to dataset
IMDB_DATADIR = os.path.join(os.path.dirname(dataset), 'aclImdb')

classes = ['pos', 'neg']
train_data = load_files(os.path.join(IMDB_DATADIR, 'train'), shuffle=True, categories=classes)
test_data = load_files(os.path.join(IMDB_DATADIR,  'test'), shuffle=False, categories=classes)

x_train = np.array(train_data.data)
y_train = np.array(train_data.target)
x_test = np.array(test_data.data)
y_test = np.array(test_data.target)

print(x_train.shape)  # (25000,)
print(y_train.shape)  # (25000, 1)
print(x_train[0][:50])  # this film was just brilliant casting

The second step is to run the TextClassifier. As a quick demo, we set epochs to 2. You can also leave the epochs unspecified for an adaptive number of epochs.

import autokeras as ak

# Initialize the text classifier.
clf = ak.TextClassifier(
    overwrite=True,
    max_trials=1)  # It only tries 1 model as a quick demo.
# Feed the text classifier with training data.
clf.fit(x_train, y_train, epochs=2)
# Predict with the best model.
predicted_y = clf.predict(x_test)
# Evaluate the best model with testing data.
print(clf.evaluate(x_test, y_test))

Validation Data

By default, AutoKeras use the last 20% of training data as validation data. As shown in the example below, you can use validation_split to specify the percentage.

clf.fit(x_train,
        y_train,
        # Split the training data and use the last 15% as validation data.
        validation_split=0.15)

You can also use your own validation set instead of splitting it from the training data with validation_data.

split = 5000
x_val = x_train[split:]
y_val = y_train[split:]
x_train = x_train[:split]
y_train = y_train[:split]
clf.fit(x_train,
        y_train,
        epochs=2,
        # Use your own validation set.
        validation_data=(x_val, y_val))

Customized Search Space

For advanced users, you may customize your search space by using AutoModel instead of TextClassifier. You can configure the TextBlock for some high-level configurations, e.g., vectorizer for the type of text vectorization method to use. You can use 'sequence', which uses TextToInteSequence to convert the words to integers and use Embedding for embedding the integer sequences, or you can use 'ngram', which uses TextToNgramVector to vectorize the sentences. You can also do not specify these arguments, which would leave the different choices to be tuned automatically. See the following example for detail.

import autokeras as ak

input_node = ak.TextInput()
output_node = ak.TextBlock(block_type='ngram')(input_node)
output_node = ak.ClassificationHead()(output_node)
clf = ak.AutoModel(
    inputs=input_node,
    outputs=output_node,
    overwrite=True,
    max_trials=1)
clf.fit(x_train, y_train, epochs=2)

The usage of AutoModel is similar to the functional API of Keras. Basically, you are building a graph, whose edges are blocks and the nodes are intermediate outputs of blocks. To add an edge from input_node to output_node with output_node = ak.[some_block]([block_args])(input_node).

You can even also use more fine grained blocks to customize the search space even further. See the following example.

import autokeras as ak

input_node = ak.TextInput()
output_node = ak.TextToIntSequence()(input_node)
output_node = ak.Embedding()(output_node)
# Use separable Conv layers in Keras.
output_node = ak.ConvBlock(separable=True)(output_node)
output_node = ak.ClassificationHead()(output_node)
clf = ak.AutoModel(
    inputs=input_node,
    outputs=output_node,
    overwrite=True,
    max_trials=1)
clf.fit(x_train, y_train, epochs=2)

Data Format

The AutoKeras TextClassifier is quite flexible for the data format.

For the text, the input data should be one-dimensional For the classification labels, AutoKeras accepts both plain labels, i.e. strings or integers, and one-hot encoded encoded labels, i.e. vectors of 0s and 1s.

We also support using tf.data.Dataset format for the training data.

train_set = tf.data.Dataset.from_tensor_slices(((x_train, ), (y_train, ))).batch(32)
test_set = tf.data.Dataset.from_tensor_slices(((x_test, ), (y_test, ))).batch(32)

clf = ak.TextClassifier(
    overwrite=True,
    max_trials=2)
# Feed the tensorflow Dataset to the classifier.
clf.fit(train_set, epochs=2)
# Predict with the best model.
predicted_y = clf.predict(test_set)
# Evaluate the best model with testing data.
print(clf.evaluate(test_set))

Reference

TextClassifier, AutoModel, TextBlock, TextToInteSequence, Embedding, TextToNgramVector, ConvBlock, TextInput, ClassificationHead.