Moonshot ML prices_to_features features and targets preparation

Before I start, I need to highlight I'm new to quantrocket but many years in AI which I'd like to leverage on this platform.

I am translating my strategy into moonshot_ML.
The premise of the ML model is to predict prices couple of days ahead based on historical prices.
preparing training dataset its many inputs (features) to many targets.

Below picture shows historical days 33 in this case turned to 33 features and 8 days to predict turned into 8 features.
The predicted days will be further transformed in the "predictions_to_signals" method.

The vanilla version (data science way) works fine with such challenge.

Below how I created MoonshotML dataset. (based on tutorial kitchensink_ml and documentation)
Screenshot 2024-02-22 at 12.12.13 AM
Here for simplicity I shrank number of features and target 6 and 4 respectively.

I attached a scratch Jupyter notebook for verification.

ML__1st.ipynb (529.2 KB)

My question is:
is creating many targets in MoonshotML the proper way to work with targets? Documentation is not clear on that.
Why my doubts?
The vanilla data science version (attached script) each row represents time-series that is shifting with every day record creating 2D array.

but in Moonshot ML prices_to_features each row is taken by symbol. Stacking them as dictionaries creates somewhat 3D array. where each feature (a day) is new slide (a dictionary containing data frame) - details in the attached script

Is the way I created many targets supported in Moonshot?
Will it cause troubles down the line for the Moonshot engine?

Currently multiple targets aren't supported, but it would be good to support that. Let me think about the best way.