Anuvada: Interpretable Models for NLP using PyTorch

One of the common criticisms of deep learning has been it's black box nature. To address this issue, researchers have developed many ways to visualise and explain the inference. Some examples would be attention in the case of RNN's, activation maps, guided back propagation and occlusion (in the case of CNN's). This library is an ongoing effort to provide a high-level access to such models relying on PyTorch.

Installing

Clone this repo and add it to your python library path.

Getting started

Importing libraries

import anuvadaimport numpy as npimport torchimport pandas as pd

from anuvada.models.classification_attention_rnn import AttentionClassifier

Creating the dataset

from anuvada.datasets.data_loader import CreateDatasetfrom anuvada.datasets.data_loader import LoadData

data = CreateDataset()

df = pd.read_csv('MovieSummaries/movie_summary_filtered.csv')

# passing only the first 512 samples, I don't have a GPU!y = list(df.Genre.values)[0:512]x = list(df.summary.values)[0:512]

x, y = data.create_dataset(x,y, folder_path='test', max_doc_tokens=500)

Loading created dataset

l = LoadData()

x, y, token2id, label2id, lengths_mask = l.load_data_from_path('test')

Change into torch vectors

x = torch.from_numpy(x)

y = torch.from_numpy(y)

Create attention classifier

acf = AttentionClassifier(vocab_size=len(token2id),embed_size=25,gru_hidden=25,n_classes=len(label2id))

loss = acf.fit(x,y, lengths_mask ,epochs=5)

Epoch 1 / 5
[========================================] 100% loss: 3.9904loss: 3.9904

Epoch 2 / 5
[========================================] 100% loss: 3.9851loss: 3.9851

Epoch 3 / 5
[========================================] 100% loss: 3.9783loss: 3.9783

Epoch 4 / 5
[========================================] 100% loss: 3.9739loss: 3.9739

Epoch 5 / 5
[========================================] 100% loss: 3.9650loss: 3.9650