Switch branches/tags
Nothing to show
Find file History

README.md

Text Classification

This repository contains end-to-end tutorial-like code samples to help solve text classification problems using machine learning.

Prerequisites

Modules

We have one module for each step in the text classification workflow.

  1. load_data - Functions to load data from four different datasets. For each of the dataset we:

    • Read the required fields (texts and labels).
    • Do any pre-processing if required. For example, make sure all label values are in range [0, num_classes-1].
    • Split the data into training and validation sets.
    • Shuffle the training data.
  2. explore_data - Helper functions to understand datasets.

  3. vectorize_data - N-gram and sequence vectorization functions.

  4. build_model - Helper functions to create multi-layer perceptron and separable convnet models.

  5. train data - Demonstrates how to use all of the above modules and train a model.

    • train_ngram_model - Trains a multi-layer perceptron model on IMDb movie reviews sentiment analysis dataset.

    • train_sequence_model - Trains a sepCNN model on Rotten Tomatoes movie reviews sentiment analysis dataset.

    • train_fine_tuned_sequence_model - Trains a sepCNN model with pre-trained embeddings that are fine-tuned on Tweet weather topic classification dataset.

    • batch_train_sequence_model - Same as train_sequence_model but here we are training data in batches.

  6. tune_model - Contains example to demonstrate how you can find the best hyper-parameter values for your model.