Adversarial Sentiment Classification

Adversarial attack on sentiment classification task.

Usage

Train Model

# datasets does not exists, build datasets from scratch
python run.py --build_data --alg CNN --outfile cnn-model.pt

# datasets already exists
python run.py --alg CNN --outfile cnn-model.pt

Modules

Build and Save Datasets

# Build and save datasets
from data_loader import DataLoader

data_loader = DataLoader()
data_loader.build_data()

Load Datasets

from data_loader import DataLoader

data_loader = DataLoader()
data_loader.load_data()

# Dataset iterator
train_iter, valid_iter = data_loader.large_train_valid_iter()

Load Trained Model

data_loader = DataLoader()
data_loader.load_data()

# This model is trained using large dataset
cnn_model = load_saved_model('CNN', 'saved_model/cnn-1.pt', data_loader)

# If model is trained using small datasets, set `small_subsets=True`
cnn_model = load_saved_model('CNN', 'saved_model/cnn-small.pt', data_loaderl, small_subsets=True)