Solid framework to make sure we're progressing in building neural network

AI & Machine learning
1

Once we've completed the neutral network in trading by Ernest chan, what we been doing is that :

  • get new features & target variable to try
  • cleanse data
  • backtest and get negative result
  • repeat

     

    We feel that we are stuck in these kind of infinite loop, and we are not making progress

     

    We tried keeping track of what are the attempts/experiments/tweaks that we did, and its just theres too much moving variables in this, ans this neural network seems like a black box the more we go in-depth in to it, that we are not making any progress

     

    Is there any tips or solid framework to make sure that we are walking to the right path that we are improving to make profitable neural network ? (better to walking the right way, than running the wrong way)

     

    Thank you
2

Hello Dwi,

You can try working on a different frequency of data. There is no rule on which features would work for a particular model, but you can try working with uncorrelated features of a weakly predictive nature. The neural network actual weights and how each feature is being used is indeed a black-box for all practical purposes as the complex relationship of different neural layers and nodes can not be easily visualised by us. The key is to use large amounts of data so that the network learns to pick up the signal amidst the noise.

Hope this was a bit helpful for you!

Thanks,

Gaurav