How to backtest a machine learning model which uses a sequence data as input?

[AboveTheHeavens]

I have an ML model which uses a 128 value period (think 128 rows) to predict 128+1's output value.

I have tried to look at documentation but from my understanding I have to input a df with open, high, low, close columns.

How would I go about passing my 128 row sequence input?

Thanks a lot!

[kernc]

Have you seen "Trading with Machine Learning" tutorial? It shows an example of a walk-forward analysis.

Basically, you pass the whole df into Backtest; then df.iloc[:128] in init() (or df.iloc[-128:] in next()) at each model training stage.

[AboveTheHeavens]

Wow thanks a lot!

[AboveTheHeavens]

I don't think this works, if I try to print print(self.data.df.shape), it gives me (2, 6), so I'm only getting 2 rows at a time in next?

[kernc]

Well, next() "builds up" data incrementally. Have you read "Trading with Machine Learning" tutorial? It contains:

    def next(self):
        # Skip the training, in-sample data
        if len(self.data) < N_TRAIN:
            return

In the same way, you can skip a few steps until sufficient data is available?