if forecast_length == 'self': about autots

[winedarksea]

I think the problem here is you are passing in df as an arg to predict (which it sees as a positional arg for forecast length).
It is understandable you are trying to use the sklearn style but because there are more options here for data import there is a separate function to do that, model.fit_data(df).
In fact, there is no need to pickle AutoTS, you can simple model.export_template("your_file.csv", n=1)

then on a new AutoTS class:

model = AutoTS(...args)
model.import_best_model("your_file.csv")
model.fit_data(df)
model.predict()