Can you use custom metrics with fit? i.e. SMAPE

[pst2154]

Is there a way to use SMAPE like in lightgbm?

lgb_ver_1 = lgb.LGBMRegressor().fit(X_train, y_train, eval_metric = 
                                   lambda y_true, y_pred: [lightgbm_smape(y_true, y_pred)])

[jmoralez]

Hey. I think you can do something like this, that gives you full control on the fit.

[pst2154]

ok, this one runs but I'm not sure if its doing it correctly as there is no print message after the fit. Why does the model not give any fit information btw?

models = {
    'forecast': lgb.LGBMRegressor(**lgb_params, objective='regression', metric = 'lightgbm_smape'),
}

[jmoralez]

What do you mean?

[pst2154]

Sorry, I think i just mean getting model metrics out after you run the fit command. There kist does not seem to be much information from the lightGBM model stored in the forecast object after you call the fit method. Not a big deal

[jmoralez]

Are you calling the MLForecast.fit method? With the example above you can train the model in any way you like, e.g. using LightGBM's train API with callbacks and such.

[pst2154]

Ya I think I had a brainless moment when I wrote the question