Retraining the autosort classifier

Autosorter retraining steps

• Grab the normalized_drive_text and youtube_metadata files from the gothrough client devices
• cd scripts && pip install -Ur requirements.txt
• rm -rf .gcache/gdrive/get_subfolders
• [next time] add logic to fix_yttranscript_cache.py to also pull metadata for all website yt links
• run fix_yttranscript_cache.py as needed to get missing transcripts
• run train script with the yt holdout to get last performance, then without to get the final model