generate-survey-comments

Generate artificial survey comments based on real data.

Steps

1. Collected sample data from multiple publicly available sources, including data.world. Shoot me a note if you would like the training data set, ~4k comments.

2. Combine and clean data (compile_data.R). Save as data/training_comments.csv

3. Save a copy of training_comments.csv to Google Drive. Location: '/content/gdrive/MyDrive/Development/seq2seqcomments/training_comments.csv'

4. Train model on Google Colab using seq2seqcomments.ipynb. Results were subpar.

5. Trained model on Google Colab using Interactive_textgenrnn_Comment_Generator_w_GPU.ipynb, by Max Woolf. Results were much better but still not entirely coherent. Trying again with 40 epochs.

6. Generate comments TBD

Standalone script

comments-not-on-github.py is a standalone Python script proof of concept using a small sample data set. Note that instead of model.predict_classes() the code below should be used when making predictions, per a Keras change in 2021.

predicted = np.argmax(model.predict(token_list), axis=-1)