All encoding updates

[VeritasJoker]

Makefile

• added run-erp
• added concat-lags (concatenate encoding results with different lags)
• added plot-old (plot comp and prod together), plot-new (plot comp and prod separately), plot-all (plot patients average together), plot-erp (plot erp results)

tfsenc_main

• added parallel when looping through all electrodes
• added detrending for each conversation signal
• added the option to z-score for each conversation signal (for erp)
• added datum modifications (for podcast glove: incorrect, correct, and gpt2-pred based on gpt2 embeddings) (for tfs: trim edges based on max lag, first n-words in utterances, single conversations)
• added model modifications (best lag)(comprehension/production train/test flip + best lag)
• dealt with different number of conversations for each electrode (load stitch index, add nan of correct size for missing electrodes, filter datum with all the conversations present for that specific electrode) (add cases for electrodes with no signal or with less than 5 conversations) (patient 7170)

tfsenc_parser

• added argument datum_mod (datum modifications) and model_mod (model modifications)
• added argument bad_convos (patient 676 blenderbot embeddings)

tfsenc_read_datum

• dealt with bad conversations (patient 676 blenderbot embeddings)

tfsenc_utils

• model modifications implementation
• pipeline implementation (pca inside folds)
• moved folds to a higher level function to accommodate model modifications
• specified cases for groupkfolds and kfolds

tfserp_main

• run erp

[hvgazula]

Quite some changes. Probably, we will have to work this out together.

[VeritasJoker]

Yes. I can clean up some of the codes later. If we want, I can clean up all the codes we used for the old significant test. That will shorten the code quite a lot.