nateemma
commented
1 year ago some answers below...
On Wed, Feb 22, 2023 at 9:15 AM nilux @.***> wrote:
@nateemma https://github.com/nateemma thank you so much for your repo - I'm working on my modified PCA strategies for a few weeks now with great results. I really enjoy diving deeper into all this.
Two questions:
-
Currently the model of PCA is stored in memory right? Wouldn't it make more sense to use a persistent storage on disk to retain models even after restart? Like it's done in freqai. Is there a way to change this to disk storage?
It sort of does this already. When I run in backtesting mode, the resulting models (one per pair) are saved to disk. In other modes, those models are reloaded and used to initialise the classification algorithms, but I then re-fit using 'live' data. I haven't tried using just the saved model, but will try that when I have some time. FYI, the underlying (sklearn) classifiers are not cumulative, so I can't incrementally train them Regarding memory, you have to have the data in memory to run the classifiers anyway so saving them to disk doesn't really help.
-
I would like to incorporate the accuracy of predictions into my risk management. Basically check if the past x predictions lead to a profitable trade. How would I approach this to get such accuracy score of the recent buy signals and recent sell signals?
If you set the flags dbg_verbose and dbg_test_classifier to True, then you will see accuracy scores (using test data, not training data). If you want to re-use that code, look at train_models(0 (in PCA.py), at the end of the function (or just search for if self.dbg_test_classifier and self.dbg_verbose:
-
If you could point me in the right direction I would appreciate that! Keep up the good work
Hope that helps
Cheers,
Phil
— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/12, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X56TYV54UYVJGEFCMTLWYZCTBANCNFSM6AAAAAAVETJGB4 . You are receiving this because you were mentioned.Message ID: @.***>
just-nilux
@nateemma thank you so much for your repo - I'm working on my modified PCA strategies for a few weeks now with great results. I really enjoy diving deeper into all this.
Two questions:
Currently the classifier models of PCA are stored in memory. Wouldn't it make more sense to use a persistent storage on disk to retain models even after restart? Like it's done in freqai. Is there a way to change this to disk storage?
I would like to incorporate the accuracy of predictions into my risk management. Basically check if the past x predictions lead to a profitable trade. How would I approach this to get such accuracy score of the recent buy signals and recent sell signals?
If you could point me in the right direction I would appreciate that! Keep up the good work