Navigate to https://www.kaggle.com/c/google-quest-challenge/data for the challenge page

1. Run cd lstm to navigate to the LSTM folder

2. Run pip3 install -r requirements.txt

3. Run python3 preprocess_data.py to preprocess the QUEST dataset

4. Run python3 train.py

5. Run python3 test.py to generate a test submission file for Kaggle

6. The final, best model will be saved in the models/ folder

Steps to obtain the training accuracy (test accuracy is displayed on Kaggle) using Spearman score -

1. cd into 'log_reg' directory - cd log_reg

2. Run pip3 install -r requirements.txt

3. We then run logistic regression to generate the features, train and save the model - python logistic_regression.py (Will take around 5 minutes)

4. Above command will display the Spearman score for each label and also the overall average Spearman score, and also generate a generate a test submission file `submission.csv' for Kaggle

The naive Bayes model requires only the numpy, pandas, re, and collections libraries to run.

1. Run cd naive-bayes to navigate to the naive Bayes folder.

2. Run python3 naive-bayes.py to execute the naive Bayes model. This will train the model on the train.csv dataset and save a file submission.csv containing the predicted labels for the test.csv dataset.

The initial code repository is located at https://github.com/ElizaLo/Question-Answering-based-on-SQuAD .

1. Run cd qa to navigate to the qa folder

2. Run pip3 install -r requirements.txt

3. Run pip3 install -r requirements.txt to install all Python3 libraries

4. Run python3 -m spacy download en to download necessary Spacy models. These should be located in your Python/Python36/Lib/site-packages/en_core_web_sm/ folder.

5. Modify data_dir, spicy_en, and glove in config.py to be the locations where you want to download SQuAD to, the location from step 3 of your Spacy models on your system, and the location of your pre-trained GloVe embeddings downloaded from https://nlp.stanford.edu/projects/glove/ and extracted to that folder.

6. Run python3 make_dataset.py to preprocess the SQuAD dataset and create the vocabulary.

7. Run python3 train.py to train and save the BiDAF model.

8. Run python3 test.py to test the BiDAF model.

Before running the below programs, please download the SQuAD1.1 dataset from https://www.wolframcloud.com/objects/d91733a5-57f5-40fe-8e09-2f5285d21fe6, create a data directory, and place the file into that directory. Ensure the file is titled SQuAD-v1.1.csv before continuing.

1. Run analysis.py to generate all plots of the QUEST datasets' features, which are placed in the plots/ directory

2. Run quest_analysis.py to generate QUEST dataset statistics.

3. Run squad_analysis.py to generate SQuAD dataset statistics.

1. cd into 'log_reg' directory - cd log_reg

2. Run logistic regression if not done above - python3 logistic_regression.py

3. Now, run the command - python3 error_analysis.py to generate the error analysis plots in the plots_error_analysis folder.

1. First we parse squad json into csv for a more readable format, run the command from the parent directory- python3 parse_squad_to_csv.py. This will generate train-v1.1.csv and dev-v1.1.csv in the squad_dataset folder.

2. cd into 'log_reg' directory - cd log_reg

3. Run the command - python3 read_and_label_squad.py. This will use the saved logistic regression models to label squad. This will take ~5 minutes.

4. The labeled csv will be generated in the squad_labelled folder, dev-v1.1_labeled.csv and train-v1.1_labeled.csv.

1. cd into 'log_reg' directory - cd log_reg

2. Run the command - python3 analysis_labeled.py

3. The plots will be generated in the plots_labeled folder.