Data split

[bujowskis]

Notes:

• keep in mind the restrictions on the data interval-duration (check external python
• 10y: 1d
• 6m: 1h
• 1m: 15min

Start-end dates:

• start date (i.e. end date of the train interval) - 04.10.2022 (the day of initial commit)

• [x] fill the above

• [x] fix start-end dates of the data split

• [x] prepare data split for one stock

• [x] fixme - trader holidays

[bujowskis]

forgot to mention

• 15min interval has duration restriction up to 60 days, so for now I'm working on 1month of 15min
• 30min interval has the same duration restrictions as 15min, so I've increased it to 1h

[bujowskis]

I've added the "naive" puller, but it needs rework - note the fixme in datasplit.py

[bujowskis]

added the remaining two fixme

[white07S]

After understanding the module of the B/S model it doesn't make sense to train a model based on a time frame rather than we should train a model based on categorized stocks.

[bujowskis]

is the trader holiday problem solved then? I've seen you @white07S add some trader calendar there

[white07S]

Yeah, it's solved and you can close this issue that trader calendar is specifically for stock data to provide accurate data in matplotlib tick module to make it understandable.

Regarding training and testing, we will be splitting stock based on categorization.

[bujowskis]

can you elaborate how this split will work?

[bujowskis]

From our talk, the split is as follows:

• no longer based on time, but on volume
• divide stocks into categories
  • large volume
  • mid volume
  • small volume
• some stocks from each category are used for test and validate, the rest is train
• note - 3 models are trained

[bujowskis]

moved the remaining things that are irrelevant to this issue per se to their own issues, this one seems to be done