My modularized neural network (NN) code for data classification. I prefer using this over canned/black box procedures. The code also runs logit on the data providing a baseline comparison of NN improvements over standard statistical methods.
Google Colab Version of the Numerai Tournament Example Code
x. 20220507. This is unlikely to work since the competition has changed its data format a few times since, and I haven't actively participated in the forecasting competition for quite a while now.
v4. 20200509. Updated and verified (via Round 211) that code works with the much larger dataset.
x. 20200418. This no longer works because Numerai has increased the dataset size from 1.4GB to 3.5GB, and Colab instances cannot handle this
increase. I now use Matlab to do the NN forecasting exclusively
v3. 20190908. Updated code for the Kazutsugi tournament.
v2. 20190615. Added commented codes on staking and submitting directly from Colab.
Numerai tournament forecast code that can be run with just an Google account. It handles everything except for uploading the resulting forecast to Numera.ai. (I commented on how to upload directly from Colab in the code if that's what you really want.) Just follow the link, "open in playground," "connect" to the Google virtual machines, and run.
The Numerai tutorials that I've come across on the web all require you to
set up Python. This runs everything on Google's virtual servers, including downloading the data directly, runs/calibrates a
model, makes a forecast, and generates the forecast csv file.
If you are interested in the tournament and just starting out, this can be a good starting point for you.