ValueError: need at least one array to concatenate When using Shin2017A dataset

Expected Result

Using Shin2017A dataset to do a CrossSession LeftRightImagery classification.

Current Result

Shin2017A-CrossSession:   0%|          | 0/29 [01:04<?, ?it/s]
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[9], line 3
      1 pipes = {}
      2 pipes["EEGNetV4"] = make_pipeline(clf)
----> 3 results = evaluation.process(pipelines=pipes)
      4 print("============results===========")
      5 print(results)

File /opt/conda/lib/python3.10/site-packages/moabb/evaluations/base.py:188, in BaseEvaluation.process(self, pipelines, param_grid, postprocess_pipeline)
    179     # (we only keep the pipeline for the first frequency band, better ideas?)
    181     results = self.evaluate(
    182         dataset,
    183         pipelines,
   (...)
    186         postprocess_pipeline=postprocess_pipeline,
    187     )
--> 188     for res in results:
    189         self.push_result(res, pipelines, process_pipeline)
    191 return self.results.to_dataframe(
    192     pipelines=pipelines, process_pipeline=process_pipeline
    193 )

File /opt/conda/lib/python3.10/site-packages/moabb/evaluations/evaluations.py:548, in CrossSessionEvaluation.evaluate(self, dataset, pipelines, param_grid, process_pipeline, postprocess_pipeline)
    542         save_model_cv(
    543             model=cvclf,
    544             save_path=model_save_path,
    545             cv_index=str(cv_ind),
    546         )
    547 else:
--> 548     result = _fit_and_score(
    549         clone(grid_clf),
    550         X,
    551         y,
    552         scorer,
    553         train,
    554         test,
    555         verbose=False,
    556         parameters=None,
    557         fit_params=None,
    558         error_score=self.error_score,
    559         return_estimator=True,
    560     )
    561     score = result["test_scores"]
    562     model_list = result["estimator"]

File /opt/conda/lib/python3.10/site-packages/sklearn/model_selection/_validation.py:686, in _fit_and_score(estimator, X, y, scorer, train, test, verbose, parameters, fit_params, return_train_score, return_parameters, return_n_test_samples, return_times, return_estimator, split_progress, candidate_progress, error_score)
    684         estimator.fit(X_train, **fit_params)
    685     else:
--> 686         estimator.fit(X_train, y_train, **fit_params)
    688 except Exception:
    689     # Note fit time as time until error
    690     fit_time = time.time() - start_time

File /opt/conda/lib/python3.10/site-packages/sklearn/pipeline.py:405, in Pipeline.fit(self, X, y, **fit_params)
    403     if self._final_estimator != "passthrough":
    404         fit_params_last_step = fit_params_steps[self.steps[-1][0]]
--> 405         self._final_estimator.fit(Xt, y, **fit_params_last_step)
    407 return self

File /opt/conda/lib/python3.10/site-packages/braindecode/eegneuralnet.py:384, in _EEGNeuralNet.fit(self, X, y, **fit_params)
    382     self._set_signal_args(X, y, classes=None)
    383     self.signal_args_set_ = True
--> 384 return super().fit(X=X, y=y, **fit_params)

File /opt/conda/lib/python3.10/site-packages/skorch/classifier.py:165, in NeuralNetClassifier.fit(self, X, y, **fit_params)
    154 """See ``NeuralNet.fit``.
    155 
    156 In contrast to ``NeuralNet.fit``, ``y`` is non-optional to
   (...)
    160 
    161 """
    162 # pylint: disable=useless-super-delegation
    163 # this is actually a pylint bug:
    164 # [https://github.com/PyCQA/pylint/issues/1085](https://github.com/PyCQA/pylint/issues/1085%3C/span%3E)
--> 165 return super(NeuralNetClassifier, self).fit(X, y, **fit_params)

File /opt/conda/lib/python3.10/site-packages/skorch/net.py:1319, in NeuralNet.fit(self, X, y, **fit_params)
   1316 if not self.warm_start or not self.initialized_:
   1317     self.initialize()
-> 1319 self.partial_fit(X, y, **fit_params)
   1320 return self

File /opt/conda/lib/python3.10/site-packages/braindecode/eegneuralnet.py:334, in _EEGNeuralNet.partial_fit(self, X, y, classes, **fit_params)
    332     self._set_signal_args(X, y, classes)
    333     self.signal_args_set_ = True
--> 334 return super().partial_fit(X=X, y=y, classes=classes, **fit_params)

File /opt/conda/lib/python3.10/site-packages/skorch/net.py:1278, in NeuralNet.partial_fit(self, X, y, classes, **fit_params)
   1276 self.notify('on_train_begin', X=X, y=y)
   1277 try:
-> 1278     self.fit_loop(X, y, **fit_params)
   1279 except KeyboardInterrupt:
   1280     pass

File /opt/conda/lib/python3.10/site-packages/skorch/net.py:1196, in NeuralNet.fit_loop(self, X, y, epochs, **fit_params)
   1190     self.run_single_epoch(iterator_train, training=True, prefix="train",
   1191                           step_fn=self.train_step, **fit_params)
   1193     self.run_single_epoch(iterator_valid, training=False, prefix="valid",
   1194                           step_fn=self.validation_step, **fit_params)
-> 1196     self.notify("on_epoch_end", **on_epoch_kwargs)
   1197 return self

File /opt/conda/lib/python3.10/site-packages/skorch/net.py:386, in NeuralNet.notify(self, method_name, **cb_kwargs)
    384 getattr(self, method_name)(self, **cb_kwargs)
    385 for _, cb in self.callbacks_:
--> 386     getattr(cb, method_name)(self, **cb_kwargs)

File /opt/conda/lib/python3.10/site-packages/skorch/callbacks/scoring.py:489, in EpochScoring.on_epoch_end(self, net, dataset_train, dataset_valid, **kwargs)
    486     return
    488 with _cache_net_forward_iter(net, self.use_caching, y_pred) as cached_net:
--> 489     current_score = self._scoring(cached_net, X_test, y_test)
    491 self._record_score(net.history, current_score)

File /opt/conda/lib/python3.10/site-packages/skorch/callbacks/scoring.py:181, in ScoringBase._scoring(self, net, X_test, y_test)
    178 """Resolve scoring and apply it to data. Use cached prediction
    179 instead of running inference again, if available."""
    180 scorer = check_scoring(net, self.scoring_)
--> 181 return scorer(net, X_test, y_test)

File /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_scorer.py:234, in _BaseScorer.__call__(self, estimator, X, y_true, sample_weight)
    211 def __call__(self, estimator, X, y_true, sample_weight=None):
    212     """Evaluate predicted target values for X relative to y_true.
    213 
    214     Parameters
   (...)
    232         Score function applied to prediction of estimator on X.
    233     """
--> 234     return self._score(
    235         partial(_cached_call, None),
    236         estimator,
    237         X,
    238         y_true,
    239         sample_weight=sample_weight,
    240     )

File /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_scorer.py:276, in _PredictScorer._score(self, method_caller, estimator, X, y_true, sample_weight)
    248 def _score(self, method_caller, estimator, X, y_true, sample_weight=None):
    249     """Evaluate predicted target values for X relative to y_true.
    250 
    251     Parameters
   (...)
    273         Score function applied to prediction of estimator on X.
    274     """
--> 276     y_pred = method_caller(estimator, "predict", X)
    277     if sample_weight is not None:
    278         return self._sign * self._score_func(
    279             y_true, y_pred, sample_weight=sample_weight, **self._kwargs
    280         )

File /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_scorer.py:73, in _cached_call(cache, estimator, method, *args, **kwargs)
     71 """Call estimator with method and args and kwargs."""
     72 if cache is None:
---> 73     return getattr(estimator, method)(*args, **kwargs)
     75 try:
     76     return cache[method]

File /opt/conda/lib/python3.10/site-packages/braindecode/classifier.py:198, in EEGClassifier.predict(self, X)
    174 def predict(self, X):
    175     """Return class labels for samples in X.
    176 
    177     Parameters
   (...)
    196 
    197     """
--> 198     return self.predict_proba(X).argmax(1)

File /opt/conda/lib/python3.10/site-packages/braindecode/classifier.py:130, in EEGClassifier.predict_proba(self, X)
     97 def predict_proba(self, X):
     98     """Return the output of the module's forward method as a numpy
     99     array. In case of cropped decoding returns averaged values for
    100     each trial.
   (...)
    128 
    129     """
--> 130     y_pred = super().predict_proba(X)
    131     # Normally, we have to average the predictions across crops/timesteps
    132     # to get one prediction per window/trial
    133     # Predictions may be already averaged in CroppedTrialEpochScoring (y_pred.shape==2).
    134     # However, when predictions are computed outside of CroppedTrialEpochScoring
    135     # we have to average predictions, hence the check if len(y_pred.shape) == 3
    136     if self.cropped and self.aggregate_predictions and len(
    137             y_pred.shape) == 3:

File /opt/conda/lib/python3.10/site-packages/skorch/classifier.py:200, in NeuralNetClassifier.predict_proba(self, X)
    168 """Where applicable, return probability estimates for
    169 samples.
    170 
   (...)
    196 
    197 """
    198 # Only the docstring changed from parent.
    199 # pylint: disable=useless-super-delegation
--> 200 return super().predict_proba(X)

File /opt/conda/lib/python3.10/site-packages/skorch/net.py:1599, in NeuralNet.predict_proba(self, X)
   1597     yp = nonlin(yp)
   1598     y_probas.append(to_numpy(yp))
-> 1599 y_proba = np.concatenate(y_probas, 0)
   1600 return y_proba

ValueError: need at least one array to concatenate

Environment

Moabb Version: 1.0.0
Braindecode: 0.8.1

Start by reproducing the Shin2017A CrossSession LeftRightImagery run, then trace CrossSessionEvaluation.evaluate in evaluations/evaluations.py through the scoring path shown in evaluations/base.py. Investigate why braindecode/skorch produces no arrays for np.concatenate during prediction. Done means the reported evaluation no longer raises this error and the behavior is covered by an appropriate regression test.