Diminishing returns of external regressors

[royrvd]

Hey Orbit team!

While trying to calculate the diminishing returns of each regressor from the demo using the following code:

def calc_regressor_diminishing_return(model, df, regressor):
    reg_max = df[regressor].max()
    reg_min = df[regressor].min()
    preds = []
    sample_df = df.iloc[[0]].copy()
    for i in np.linspace(reg_min, reg_max, num=50):
        sample_df[regressor] = i
        pred = np.exp(model.predict(sample_df)["prediction"].values[0])
        preds.append(pred)
    return preds

def plot_diminishing_returns(model, df, regressors):
    ncols = 2
    nrows = int(np.ceil(len(regressors) / 2))
    fig, axes = plt.subplots(figsize=(15, 15), ncols=ncols, nrows=nrows)
    for ax, reg in zip(axes.flatten(), regressors):
        diminishing_returns = calc_regressor_diminishing_return(model, df, reg)
        ax.plot(diminishing_returns)
        ax.set_title(reg)
    return fig

plot_diminishing_returns(ktr, rw_data, regressor_col)

I got non-monotonic curves. Does that make sense or could that be an error?

[edwinnglabs]

Here is my answer with the similar question on the issues tab