Update evaluation to use new splitters and include updates

docs/source/whats_new.rst

@@ -42,6 +42,7 @@ Enhancements
 - Adding :func:`moabb.analysis.plotting.dataset_bubble_plot` plus the corresponding tutorial (:gh:`753` by `Pierre Guetschel`_)
 - Adding :func:`moabb.datasets.utils.plot_all_datasets` and update the tutorial (:gh:`758` by `Pierre Guetschel`_)
 - Improve the dataset model cards in each API page (:gh:`765` by `Pierre Guetschel`_)
+- Refactor :class:`moabb.evaluation.CrossSessionEvaluation`, :class:`moabb.evaluation.CrossSubjectEvaluation` and  :class:`moabb.evaluation.WithinSessionEvaluation` to use the new splitter classes (:gh:`769` by `Bruno Aristimunha`_)
 - Adding tutorial on using mne-features (:gh:`762` by `Alexander de Ranitz`_, `Luuk Neervens`_, `Charlynn van Osch`_ and `Bruno Aristimunha`_)
 - Creating tutorial to expose the pre-processing steps (:gh:`771` by `Bruno Aristimunha`_)
 - Add function to auto-generate tables for the paper results documentation page (:gh:`785` by `Lucas Heck`_)

examples/advanced_examples/plot_grid_search_withinsession.py

@@ -9,7 +9,6 @@
 """
 
 import os
-from pickle import load
 
 import matplotlib.pyplot as plt
 import seaborn as sns
@@ -132,44 +131,3 @@
 )
 sns.pointplot(data=result, y="score", x="pipeline", ax=axes, palette="Set1")
 axes.set_ylabel("ROC AUC")
-
-##########################################################
-# Load Best Model Parameter
-# -------------------------
-# The best model are automatically saved in a pickle file, in the
-# results directory. It is possible to load those model for each
-# dataset, subject and session. Here, we could see that the grid
-# search found a l1_ratio that is different from the baseline
-# value.
-
-with open(
-    "./Results/Models_WithinSession/BNCI2014-001/1/1test/GridSearchEN/fitted_model_best.pkl",
-    "rb",
-) as pickle_file:
-    GridSearchEN_Session_E = load(pickle_file)
-
-print(
-    "Best Parameter l1_ratio Session_E GridSearchEN ",
-    GridSearchEN_Session_E.best_params_["LogistReg__l1_ratio"],
-)
-
-print(
-    "Best Parameter l1_ratio Session_E VanillaEN: ",
-    pipelines["VanillaEN"].steps[2][1].l1_ratio,
-)
-
-with open(
-    "./Results/Models_WithinSession/BNCI2014-001/1/0train/GridSearchEN/fitted_model_best.pkl",
-    "rb",
-) as pickle_file:
-    GridSearchEN_Session_T = load(pickle_file)
-
-print(
-    "Best Parameter l1_ratio Session_T GridSearchEN ",
-    GridSearchEN_Session_T.best_params_["LogistReg__l1_ratio"],
-)
-
-print(
-    "Best Parameter l1_ratio Session_T VanillaEN: ",
-    pipelines["VanillaEN"].steps[2][1].l1_ratio,
-)

moabb/evaluations/__init__.py

@@ -10,4 +10,4 @@
     WithinSessionEvaluation,
 )
 from .splitters import CrossSessionSplitter, CrossSubjectSplitter, WithinSessionSplitter
-from .utils import create_save_path, save_model_cv, save_model_list
+from .utils import _create_save_path, _save_model_cv

moabb/evaluations/base.py

@@ -3,29 +3,23 @@
 from warnings import warn
 
 import pandas as pd
+from joblib import Parallel, delayed
 from sklearn.base import BaseEstimator
-from sklearn.model_selection import GridSearchCV
 
 from moabb.analysis import Results
 from moabb.datasets.base import BaseDataset
-from moabb.evaluations.utils import _convert_sklearn_params_to_optuna
+from moabb.evaluations.utils import (
+    _convert_sklearn_params_to_optuna,
+    check_search_available,
+)
 from moabb.paradigms.base import BaseParadigm
 
 
+search_methods, optuna_available = check_search_available()
+
 log = logging.getLogger(__name__)
 
 # Making the optuna soft dependency
-try:
-    from optuna.integration import OptunaSearchCV
-
-    optuna_available = True
-except ImportError:
-    optuna_available = False
-
-if optuna_available:
-    search_methods = {"grid": GridSearchCV, "optuna": OptunaSearchCV}
-else:
-    search_methods = {"grid": GridSearchCV}
 
 
 class BaseEvaluation(ABC):
@@ -83,6 +77,8 @@ class BaseEvaluation(ABC):
        optuna, time_out parameters.
     """
 
+    search = False
+
     def __init__(
         self,
         paradigm,
@@ -201,7 +197,6 @@ def process(self, pipelines, param_grid=None, postprocess_pipeline=None):
             This pipeline must be "fixed" because it will not be trained,
             i.e. no call to ``fit`` will be made.
 
-
         Returns
         -------
         results: pd.DataFrame
@@ -216,26 +211,44 @@ def process(self, pipelines, param_grid=None, postprocess_pipeline=None):
             if not (isinstance(pipeline, BaseEstimator)):
                 raise (ValueError("pipelines must only contains Pipelines " "instance"))
 
-        res_per_db = []
-        for dataset in self.datasets:
-            log.info("Processing dataset: {}".format(dataset.code))
-            process_pipeline = self.paradigm.make_process_pipelines(
+        # Prepare dataset processing parameters
+        processing_params = [
+            (
                 dataset,
-                return_epochs=self.return_epochs,
-                return_raws=self.return_raws,
-                postprocess_pipeline=postprocess_pipeline,
-            )[0]
-            # (we only keep the pipeline for the first frequency band, better ideas?)
-
-            results = self.evaluate(
-                dataset,
-                pipelines,
-                param_grid=param_grid,
-                process_pipeline=process_pipeline,
-                postprocess_pipeline=postprocess_pipeline,
+                self.paradigm.make_process_pipelines(
+                    dataset,
+                    return_epochs=self.return_epochs,
+                    return_raws=self.return_raws,
+                    postprocess_pipeline=postprocess_pipeline,
+                )[0],
             )
+            for dataset in self.datasets
+        ]
+
+        # Parallel processing...
+        parallel_results = Parallel(n_jobs=self.n_jobs)(
+            delayed(
+                lambda d, p: list(
+                    self.evaluate(
+                        d,
+                        pipelines,
+                        param_grid=param_grid,
+                        process_pipeline=p,
+                        postprocess_pipeline=postprocess_pipeline,
+                    )
+                )
+            )(dataset, process_pipeline)
+            for dataset, process_pipeline in processing_params
+        )
+
+        res_per_db = []
+        # Process results in order
+        for (dataset, process_pipeline), results in zip(
+            processing_params, parallel_results
+        ):
             for res in results:
                 self.push_result(res, pipelines, process_pipeline)
+
             res_per_db.append(
                 self.results.to_dataframe(
                     pipelines=pipelines, process_pipeline=process_pipeline
@@ -316,9 +329,12 @@ def _grid_search(self, param_grid, name, grid_clf, inner_cv):
                     return_train_score=True,
                     **extra_params,
                 )
+                self.search = True
                 return search
             else:
+                self.search = True
                 return grid_clf
 
         else:
+            self.search = False
             return grid_clf