Hello again team, any one know how to add a custom loss metric in to the AutoMLForecast class? I'm currently working in retail and I need to apply a custom metric in to my CV, any help is welcome :D

Hey. You need to implement a function that takes the out of sample df and the training dataframe and returns a single float. The first df will have the predictions in a column named

model

, so you can do for example:

from utilsforecast.losses import rmse

def my_loss(df, train_df):
    return rmse(df, models=["model"])["model"].mean()

So if I want to implement a custom code metri, it would look something like this?

def my_loss(df, train_df):
    # Calculate RMSE
    rmse_value = np.sqrt(((train_df['y'] - df['model']) ** 2).mean())
    return rmse_value

the

y

has to come from

df

as well, e.g.

rmse_value = np.sqrt((df['y'] - df['model']) ** 2).mean())

Ohh I get it now, thanks a lot

The train_df is there in case you want a metric that depends on the training set like mase