✅Allow specifying the logs to be trained for motion extraction

relevance

Currently, "text files written out in advance with one line and one remark" are used as the learning target.

This was good for learning before operation, but after actual operation, "This is how I reacted to this input" triggers re-learning, so you will want to use inputs that are not in this file.

code:ugoki/user.py

def _talkid_to_logs(talkid):

data = json.load(open(f"{_HERE}/../../logs/cache/{talkid}.json"))

if user == 1:

yield text

def get_unknowns():

logs = _talkid_to_logs("alUKPkDIxF4BN18b4Yoe")

for text in logs:

for case in text_to_cases(text):

yield case

You might think it would be a good idea to "target all logs" (and I actually did it a long time ago), but it's not good because there are some strange logs that keep posting meaningless strings of text and abusing bots.

---