Where is the file train_encoded_gpt2.pkl created?

[ryanmiller1988]

Would like to run train.py using the example provided in README.md:

python train.py --data_dir datadir --output_dir savedir --experiment_name pmfull --accum_iter 4 --n_batch 64 --p 90 --num_epochs 10 --use_model plotmachines --use_neighbor_feat --use_discourse

This produces a file not found error, the code is looking for train_encoded_gpt2.pkl in the data directory.

How or where does this pickle file first get created?

Turning off use_neighbor_feat and/or use_discourse has no effect, same error.

[lourdescrivelli]

From my understanding, the pkl file should be created by using the tfmclassifier function, though I could not find in their code a call to the function that creates the pkl file. Were you able to get the full model running?

[audreycs]

This code can generate .pkl files successfully.