Question: Ranges Variable and Walk Forward Optimization

To Whom It May Concern,

I see in https://github.com/jmrichardson/tuneta/blob/main/examples/tune_all.py, that you are tuning all indicator over a period of 10 years of daily data given the next day’s returns. What does ranges=[(2, 30)] do in this instance? I see that ranges is defined here, https://github.com/jmrichardson/tuneta/blob/main/tuneta/tune_ta.py#L44. It specifies that it is a parameter search space. It appears to be used to define a low and high here, https://github.com/jmrichardson/tuneta/blob/main/tuneta/tune_ta.py#L82, which then appears to be used as a parameter in an Optuna trial here, https://github.com/jmrichardson/tuneta/blob/main/tuneta/tune_ta.py#L125. Unfortunately, it’s still not clear to me what this parameter does. My apologies for my ignorance.

I’m using pandas-ta on minute data. So, I need to adjust the default setting for most of indicators I’m using. I had hope to use this library to get these settings using a walk forward optimization strategy, which ensures that I’m not overfitting my data and/or getting stuck in a local minima. Just for completeness as I wasn’t aware of this a few months back, but a walk forward optimization is defined here, https://en.wikipedia.org/wiki/Walk_forward_optimization and https://www.youtube.com/watch?v=GowmmrSMw9I, and shown in code https://github.com/polakowo/vectorbt/blob/master/examples/WalkForwardOptimization.ipynb on VectorBT, as well as visualized here, http://www.adaptivetradingsystems.com/blog/modeling_sofware/walk-forward-simulations-in-synergy/.

Is it possible to use tuneta to adjust the setting for pandas-ta indicators in this way. If not, could we discuss how to make this possible?

Thanks for the hard work on this library, as well as your time and attention to this matter. I hope that this message finds you well and that you have a great week. God bless.

Very Respectfully, CMobley7

[CMobley7]

@jmrichardson , thanks again for your quick and thorough responses! I might have some time this upcoming weekend or next to submit a PR for this.

[jmrichardson]

@CMobley7

It looks like @wouldayajustlookatit repo is based on an earlier version prior to a major code update to support different enhancements such as distance correlation, multiple equities, cluster based selection of parameters, etc. Unfortunately, I can’t merge the updates because the code base is significantly different between both. However, if you see any updates that make sense, feel free to add/test them with your PR. Happy to merge any improvements!