Skip to content

Evaluation of Natural language Inference datasets on State of the art models and their combinations and ensembles.

Notifications You must be signed in to change notification settings

Ved2000/NLI-datasets-evaluation

Repository files navigation

NLI-dataset-evaluation

Evaluation of Natural language Inference datasets on State of the art models and their combinations and ensembles. Most of the times models work good only on some datasets but t to be appplicable in real life it needs to be give a decent performance on others too. This repository is dedicted on training the models on one datasets and testing on others.

Data

For downloading and obtaining formatted SNLI and MultiNLI datasets run the following:*

chmod u+r+x get_transfer_data.bash
./get_transfer_data.bash

Hans.txt is HANS dataset in Json format, simply convert it to a pandas dataframe and use it accordingly:*

import pandas as pd
read_file = pd.read_csv('path_to_hans.txt',delimiter='\t')

Allen NLP model

Allen NLP model is trained on SNLI datasets to test it on Hans and MLNI datasets run the notebooks added accordingly.

Infersent Model

InferSent is a sentence embeddings method that provides semantic representations for English sentences. It is trained on natural language inference data and generalizes well to many different tasks. To test it on MultiNLI dataset and HANS replace SNLI dataset files with corresponding files and test it either using notebook added or following instructions on their repository.

About

Evaluation of Natural language Inference datasets on State of the art models and their combinations and ensembles.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published