When using machine learning models, how can I use my own INDICATOR data that I have already computed, instead of computing it in real time？

[yys123]

When using machine learning models, how can I use my own indicator data that I have already computed, instead of computing it in pybroker？I want to use dataframe form to import indicator data directly.

[edtechre]

Hi @yys123,

You can add the indicator data as columns to your DataFrame and register them using pybroker.register_columns(). For an example, refer to the rsi column in the Creating a Custom Data Source notebook.

[yys123]

Thank you for your reply.
I have done as you said, if I run this code, it has no problem:

config = StrategyConfig(bootstrap_sample_size=100)
strategy = Strategy(YFinance(), '3/1/2017', '3/1/2022', config)
strategy.add_execution(None, ['NVDA', 'AMD'], models=model_slr)
strategy.backtest(train_size=0.5)

but when running the following code, it reports an error:

config = StrategyConfig(bootstrap_sample_size=100)
strategy = Strategy(YFinance(), '3/1/2017', '3/1/2022', config)
strategy.add_execution(hold_long, ['NVDA', 'AMD'], models=model_slr)
strategy.backtest(train_size=0.5)
error：

ValueError: at least one array or dtype is required

Besides, the 'strategy.walkforward' function can not work in any case if I used my own indicator data.

[edtechre]

The error you pasted sounds like it's coming from Scikit learn. Normally, indicators computed in PyBroker are used as columns for the input data passed to model#predict(). Since you are using custom columns, you will need to set input_data_fn to pass your custom columns to your model when making predictions in hold_long e.g.

pybroker.register_columns('ind1', 'ind2', 'ind3')

def input_data_fn(df):
    return df[['ind1', 'ind2', 'ind3']]  # Passes ind1, ind2, ind3 columns to model#predict()

model = pybroker.model('my_model', train_fn, input_data_fn=input_data_fn)

You were seeing the error above only for hold_long because predictions are only computed when there is an execution function set.

Besides, the 'strategy.walkforward' function can not work in any case if I used my own indicator data.

Calling strategy.walkforward will work with your own indicator data by making the change I described.

[yys123]

Thank you for your answer, it helped me solve this problem!

[edtechre]

Note, as of v1.1.12, you can now return a list of column names to use as input from your train function:

def train_fn(symbol, train_data, test_data):
    ...
    return model, ['ind1', 'ind2', 'ind3']

You will still need to register ind1, ind2, ind3 using pybroker.register_columns.