Skip to content

Latest commit

 

History

History
37 lines (22 loc) · 1.31 KB

README.md

File metadata and controls

37 lines (22 loc) · 1.31 KB

Sentence Sentiment Classifier

This example builds sentence convolutional classifier, and trains on SST data. The example configuration config_kim.py corresponds to the paper (Kim) Convolutional Neural Networks for Sentence Classification.

The example shows:

  • Contruction of simple model, involving the Embedder and Conv1DClassifier.
  • Use of Texar MultiAlignedData to read parallel text and label data.

Usage

Use the following command to download and prepare the SST binary data:

python sst_data_preprocessor.py --data-path ./data

Here

  • --data-path specifies the directory to store the SST data. If the data files do not exist, the program will automatically download, extract, and pre-process the data.

The following command trains the model with Kim's configuration:

python classifier_main.py --config config_kim

Here:

  • --config specifies the config file to use. E.g., the above use the configuration defined in config_kim.py

The model will begin training and evaluating on the validation data, and will evaluate on the test data after every epoch if a valid accuracy is obtained.

Results

The model achieves around 83% test set accuracy.