Back to Community
Google Tensorflow Package addition

Hello,

Just wondering if Quantopian is planning on incorporating the tensorflow package anytime soon?

I think this inclusion will greatly accelerate the machine learning/deep learning applications and attempts on Quantopian, especially now with minute-level data.

I would ask for Theano, but probably Google backed Tensorflow is more "legitimate" and "safer".

15 responses

tensorflow, please.

A better package will be keras, which supports both Tensorflow or Theano backend. Please add support to keras.

I agree, keras is easier to use

+1 Tensorflow (and update sklearn)

+1 for Keras, and for auto-sklearn! ☺

+1 tensorflow would be amazing

+1 for keras

+1 for TensorFlow

+1 for keras, in the short term if not training, atleastallow the package so we are able to load pre-trained models and use that to predict and run back tests or trade live.

+1 Tensorflow and Keras

+1 tensorflow, theano, keras, hyperopt (https://github.com/hyperopt/hyperopt, for Bayesian parameter search )

+1 for hyperopt too
Speaking about parameter optimisation, TPOT seems to be an excellent tool as well - IDK how it compares with auto-sklearn (no benchmarks have been made yet), nevertheless, it is definitely a great alternative, not to mention that we can experiment with both and see which one is better. ;-)
Unfortunately, it seems to me that Q's sysadmins do not give a f♂ck about our wishes→ we have been requesting some of these libraries since 2015, yet no progress has apparently been made (if it was, why would they not have told us?) !
Given the fact that we have to wait months even to a simple charting/data delivery bugfix (whenever you open the live trading dashboard while market is closed, the platform does not load last day data in charts, x.1% in the Live trading equity chart always 'untruncates' to x.09%, etc...), I am afraid that I won't see these libraries working before becoming elderly, despite being 21 yo

Tensorflow is probably not cost efficient for Q, but maybe a +1 for xgboost...on the contrary, maybe allowing fetcher in the contest would be best. Anything that allows signals from more computationally intensive algorithms would bring forth deep learning into the competition.

++1 TensorFlow please

Quantopian should work with https://www.floydhub.com/ I have used them when i did the Udacity Deep Leanring cource and having Floydhub integrated in a smart way with Quantopian would be super nice!