https://nixtla.github.io/neuralforecast/examples/intermittentdata.html can any one told me if we are just split the data to validation and train or we need the test data in tunning step and what is the specific size for horizon and validation and does the horizan involve in fit fuction or have impact becaise i am getting error here when put the horizan zero i am just trying to know how can i choose the size for the validation and the horizon do not get error in the tensor dimensions note i am just using fit i dont use cross validation

The .fit method uses ALL the data available to train the models. Horizon 0 raises an error because it means that nothing should be forecasted. If you want to evaluate the accuracy of your forecast you can manually split the data. Something like

Y_df = AirPassengersDF.copy()
mean = Y_df[Y_df.ds<='1959-12-31']['y'].mean()
std = Y_df[Y_df.ds<='1959-12-31']['y'].std()

Y_train_df = Y_df[Y_df.ds<='1959-12-31'] # 132 train
Y_test_df = Y_df[Y_df.ds>'1959-12-31']   # 12 test

However, we rather recommend using the the crossvalidation function to evaluate perfomance across different windowos. BTW: this should be in #neural-forecast

i know when make the horizan 0 it means no things to forecasting but i guess i didn't explain my question in good way