[BUG] ShapExplainer + static covariates

[alexanderlange53]

Describe the bug
Hi all,
the ShapExplainer is throwing an error when used with static covariates:
The number of features in data (10) is not the same as it was in training data (11)

To Reproduce

import numpy as np
import pandas as pd
from darts.explainability import ShapExplainer

from darts import TimeSeries
from darts.models import LightGBMModel
from darts.utils import timeseries_generation as tg

period = 20
sine_series = tg.sine_timeseries(
    length=4 * period, value_frequency=1 / period, column_name="series", freq="h"
)

sine_vals = sine_series.values()
linear_vals = np.expand_dims(np.linspace(1, -1, num=19), -1)

sine_vals[21:40] = linear_vals
sine_vals[61:80] = linear_vals
irregular_series = TimeSeries.from_times_and_values(
    values=sine_vals, times=sine_series.time_index, columns=["series"]
)

def get_model_params_2():
    """helper function that generates model parameters"""
    return {
        "lags": int(period / 2),
        "output_chunk_length": int(period / 2),
        "add_encoders": {
            "datetime_attribute": {"future": ["hour"]}
        },  
    }

sine_series_st_bin = sine_series.with_static_covariates(
    pd.DataFrame(data={"curve_type": [1]})
)
irregular_series_st_bin = irregular_series.with_static_covariates(
    pd.DataFrame(data={"curve_type": [0]})
)

train_series = [sine_series_st_bin, irregular_series_st_bin]
for series in train_series:
    print(series.static_covariates)

model =  LightGBMModel(**get_model_params_2())

model.fit(train_series)
explainer = ShapExplainer(model, background_series=train_series)
explainer.summary_plot()

System:

• Python version: [3.8.10]
• darts version [0.23]

Additional context
The error message says that the number of features in the data passed to the ShapExplainer does not match the number of features in the data used to train the model, when in fact it does. The example is with LightGBM, but I get the same error with other models.

[dennisbader]

@alexanderlange53, thanks for raising that. Unfortunately, ShapExplainer does not yet support static covariates.
Removing the static covariates from your input series when training the model should work.

I opened ticket #1457 for that.

[madtoinou]

Fixed by #1803.