All Projects → EdGENetworks → anuvada

EdGENetworks / anuvada

Licence: other
Interpretable Models for NLP using PyTorch

Programming Languages

python
139335 projects - #7 most used programming language

Projects that are alternatives of or similar to anuvada

query-wellformedness
25,100 queries from the Paralex corpus (Fader et al., 2013) annotated with human ratings of whether they are well-formed natural language questions.
Stars: ✭ 80 (-21.57%)
Mutual labels:  nlp-machine-learning
topic modelling financial news
Topic modelling on financial news with Natural Language Processing
Stars: ✭ 51 (-50%)
Mutual labels:  nlp-machine-learning
Deception-Detection-on-Amazon-reviews-dataset
A SVM model that classifies the reviews as real or fake. Used both the review text and the additional features contained in the data set to build a model that predicted with over 85% accuracy without using any deep learning techniques.
Stars: ✭ 42 (-58.82%)
Mutual labels:  nlp-machine-learning
cross target stance classification
Keras implementation for CrossNet in paper "Cross-Target Stance Classification with Self-Attention Networks"
Stars: ✭ 21 (-79.41%)
Mutual labels:  nlp-machine-learning
scicle-stopclickbait
Userscript that changes Clickbait headlines by headlines more honest to the news it links to.
Stars: ✭ 16 (-84.31%)
Mutual labels:  nlp-machine-learning
Entity Embedding
Reference implementation of the paper "Word Embeddings for Entity-annotated Texts"
Stars: ✭ 19 (-81.37%)
Mutual labels:  nlp-machine-learning
schrutepy
The Entire Transcript from the Office in Tidy Format
Stars: ✭ 22 (-78.43%)
Mutual labels:  nlp-machine-learning
Quora QuestionPairs DL
Kaggle Competition: Using deep learning to solve quora's question pairs problem
Stars: ✭ 54 (-47.06%)
Mutual labels:  nlp-machine-learning
Sumrized
Automatic Text Summarization (English/Arabic).
Stars: ✭ 37 (-63.73%)
Mutual labels:  nlp-machine-learning
deep-semantic-code-search
Deep Semantic Code Search aims to explore a joint embedding space for code and description vectors and then use it for a code search application
Stars: ✭ 63 (-38.24%)
Mutual labels:  nlp-machine-learning
SENet-for-Weakly-Supervised-Relation-Extraction
No description or website provided.
Stars: ✭ 39 (-61.76%)
Mutual labels:  nlp-machine-learning
fake-news
This is a further development of the kdnuggets article on fake news classification by George McIntyre
Stars: ✭ 15 (-85.29%)
Mutual labels:  nlp-machine-learning
DeepLearningReading
Deep Learning and Machine Learning mini-projects. Current Project: Deepmind Attentive Reader (rc-data)
Stars: ✭ 78 (-23.53%)
Mutual labels:  nlp-machine-learning
awesome-yoruba-nlp
📖 A curated list of resources dedicated to Natural Language Processing (NLP) in the Yoruba Language.
Stars: ✭ 21 (-79.41%)
Mutual labels:  nlp-machine-learning
Engine
The Centrifuge process, filter and saves the relevant documents as recommendations to the relevant users
Stars: ✭ 20 (-80.39%)
Mutual labels:  nlp-machine-learning
merkalysis
A marketing tool that helps you to market your products using organic marketing. This tool can potentially save you 1000s of dollars every year. The tool predicts the reach of your posts on social media and also suggests you hashtags for captions in such a way that it increases your reach.
Stars: ✭ 28 (-72.55%)
Mutual labels:  nlp-machine-learning
empythy
Automated NLP sentiment predictions- batteries included, or use your own data
Stars: ✭ 17 (-83.33%)
Mutual labels:  nlp-machine-learning
Very-deep-cnn-tensorflow
Very deep CNN for text classification
Stars: ✭ 18 (-82.35%)
Mutual labels:  nlp-machine-learning
AI-Sentiment-Analysis-on-IMDB-Dataset
Sentiment Analysis using Stochastic Gradient Descent on 50,000 Movie Reviews Compiled from the IMDB Dataset
Stars: ✭ 55 (-46.08%)
Mutual labels:  nlp-machine-learning
Machine-Learning-Models
In This repository I made some simple to complex methods in machine learning. Here I try to build template style code.
Stars: ✭ 30 (-70.59%)
Mutual labels:  nlp-machine-learning

Anuvada: Interpretable Models for NLP using PyTorch

So, you want to know why your classifier arrived at a particular decision or why your flashy new deep learning classification model is not performing in the way which you would want it to perform? Or there could be bias in your dataset towards a particular class and you want to understand if there are any such edge cases.

One of the common criticisms of deep learning has been it's black box nature (life itself is a big black box, not at all interpretable, don't even ask me about love). To address this issue, researchers have developed many ways to visualise and explain the inference. It is not necessary that a model has to be explainable, but when important decisions like which jobs to recommend to a person or whether to give a person loan are being made, it would be helpful to cross-check the model's claims. In such domains, self-explainable models are necessary.

This library is an ongoing effort to provide a high-level access to such models by building on top of PyTorch.

Here is what you can expect to visualize from a trained model.

Note: This model is a convolutional neural network trained on IMDB sentiment analysis dataset. I trained the model using SGD till validation loss stopped improving. Here is sensitivity analysis on some sample inputs. You can find more details about training the model in the Jupyter notebooks from the examples directory.

Positive review

alt text

Negative review

alt text

Installing

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

Requirements

To do list

  • Implement Attention with RNN
  • Implement Attention Visualisation
  • Implement working Fit Module
  • Implement support for masking gradients in RNN (Working now!)
  • Implement a generic data set loader
  • Implement CNN Classifier * (pending) with feature map visualisation
  • Sensitivity analysis for generic model (CNN done, pending RNN)
  • Implement support for validation split
  • Separate modules for CNN and RNN
  • Implement GPU Support
  • Implement support for custom loss function, learning rate tuning, custom metrics
  • Support for loading and saving models
  • Write documentation
  • Clean Code
  • Implement basic validation measures like accuracy

Acknowledgments

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].