multi-class-text-classification-cnn-rnn

Project: Classify Kaggle San Francisco Crime Description

Highlights

• Multi-class text classification (sentence classification) problem.
• Classify Kaggle San Francisco Crime Descript into 39 Category labels.
• Hybrid TextCNN + GRU model implemented with TensorFlow 2.x Keras.

Data: Kaggle San Francisco Crime

• Input: Descript
• Output: Category

Examples:

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Training data is included at ./data/train.csv.zip. Prediction sample data is at ./data/small_samples.csv.

Train

python3 train.py ./data/train.csv.zip ./training_config.json

Artifacts are written to ./trained_results_/:

• saved_model/ — exported SavedModel for inference
• best_model.keras — best validation checkpoint (primary load path for predict)
• words_index.json — vocabulary mapping
• labels.json — class labels
• trained_parameters.json — hyperparameters and sequence length

Predict

python3 predict.py ./trained_results_/ ./data/small_samples.csv

Predictions are saved to ./predicted_results_/predictions_all.csv.

Reference

• Implement a CNN for text classification in TensorFlow