Create preprocessing.py (#238)

* Create preprocessing.py

* add comments to preprocessing.py

* [pre-commit.ci] auto fixes from pre-commit.com hooks

* use ndscaler in plot_financial_data

* typo

* missing import

* [pre-commit.ci] auto fixes from pre-commit.com hooks

* tests

* [pre-commit.ci] auto fixes from pre-commit.com hooks

* missing import

* pytest import not required

* api doc

* correct docstring

* Change NDRobustScaler -> NdRobustScaler

* Update pyriemann_qiskit/utils/preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

* Update pyriemann_qiskit/utils/preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

* fix: syntax error

* [pre-commit.ci] auto fixes from pre-commit.com hooks

* Update pyriemann_qiskit/utils/preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

* Update pyriemann_qiskit/utils/preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

* Update pyriemann_qiskit/utils/preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

* Update tests/test_preprocessing.py

Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>

doc/api.rst

@@ -54,6 +54,16 @@ Hyper-parameters generation
     get_spsa
     get_spsa_parameters
 
+Preprocessing
+~~~~~~~~~~~~~~~~~~~~~~~~~~~
+.. _preprocessing_api:
+.. currentmodule:: pyriemann_qiskit.utils.preprocessing
+
+.. autosummary::
+    :toctree: generated/
+
+    NdRobustScaler
+
 Filtering
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 .. _filtering_api:

examples/other_datasets/plot_financial_data.py

@@ -35,7 +35,7 @@
 from sklearn.model_selection import HalvingGridSearchCV
 from sklearn.pipeline import make_pipeline
 from sklearn.model_selection import train_test_split
-from sklearn.preprocessing import RobustScaler, LabelEncoder
+from sklearn.preprocessing import LabelEncoder
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.svm import SVC
 from sklearn.metrics import balanced_accuracy_score
@@ -44,6 +44,7 @@
 from pyriemann.estimation import XdawnCovariances
 from pyriemann.utils.viz import plot_waveforms
 from pyriemann_qiskit.classification import QuanticSVM
+from pyriemann_qiskit.utils.preprocessing import NdRobustScaler
 from matplotlib import pyplot as plt
 import warnings
 import pandas as pd
@@ -68,7 +69,7 @@
 
 def plot_ERP(X, title, n=10, ylim=None, add_digest=False):
     epochs = ToEpochs(n=n).transform(X)
-    reduced_centered_epochs = NDRobustScaler().fit_transform(epochs)
+    reduced_centered_epochs = NdRobustScaler().fit_transform(epochs)
     fig = plot_waveforms(reduced_centered_epochs, "mean+/-std")
     fig.axes[0].set_title(f"{title} ({len(X)})")
     if ylim is None:
@@ -85,7 +86,7 @@ def plot_ERP(X, title, n=10, ylim=None, add_digest=False):
 
 
 def merge_2axes(fig1, fig2, file_name1="f1.png", file_name2="f2.png"):
-    # Modified from [5]
+    # Modified from [4]
     fig1.savefig(file_name1)
     fig2.savefig(file_name2)
     plt.close(fig1)
@@ -174,9 +175,6 @@ def plot_ERPs(X, y, n=10, label0="Fraud", label1="Genuine"):
     lambda x: 2023 - x.year
 )
 
-# features["PK_TSINSERCION"] = pd.to_datetime(features["PK_TSINSERCION"])
-# features["PK_TSINSERCION"] = features["PK_TSINSERCION"].apply(lambda x: x.value)
-
 # Let's encode our categorical variables (LabelEncoding):
 # features["IP_TERMINAL"] = features["IP_TERMINAL"].astype("category").cat.codes
 le = LabelEncoder()
@@ -216,28 +214,6 @@ def transform(self, X):
         return all_epochs
 
 
-# Apply one scaler by channel:
-# See Stackoverflow link for more details [4]
-class NDRobustScaler(TransformerMixin):
-    def __init__(self):
-        self._scalers = []
-
-    def fit(self, X, y=None, **kwargs):
-        _, n_channels, _ = X.shape
-        self._scalers = []
-        for i in range(n_channels):
-            scaler = RobustScaler()
-            scaler.fit(X[:, i, :])
-            self._scalers.append(scaler)
-        return self
-
-    def transform(self, X, **kwargs):
-        n_channels = len(self._scalers)
-        for i in range(n_channels):
-            X[:, i, :] = self._scalers[i].transform(X[:, i, :])
-        return X
-
-
 def slim(x, keep_diagonal=True):
     # Vectorize covariance matrices by removing redundant information.
     length = len(x) // 2
@@ -285,7 +261,7 @@ def transform(self, X):
 # then adds at the end a classical SVM
 pipe = make_pipeline(
     ToEpochs(n=10),
-    NDRobustScaler(),
+    NdRobustScaler(),
     XdawnCovariances(nfilter=1),
     OptionalWhitening(process=True, n_components=4),
     SlimVector(keep_diagonal=True),
@@ -491,6 +467,4 @@ def predict(self, X):
 #         https://zenodo.org/records/7418458
 # .. [3] https://pyriemann.readthedocs.io/en/latest/auto_examples/ERP/plot_ERP.html
 #
-#    [4] https://stackoverflow.com/questions/50125844/how-to-standard-scale-a-3d-matrix
-#
-#    [5] https://stackoverflow.com/questions/16748577
+# .. [4] https://stackoverflow.com/questions/16748577

pyriemann_qiskit/utils/__init__.py

@@ -1,4 +1,4 @@
-from . import hyper_params_factory, filtering
+from . import hyper_params_factory, filtering, preprocessing
 from .quantum_provider import get_provider, get_devices, get_simulator
 from .math import cov_to_corr_matrix
 from .docplex import (
@@ -23,6 +23,7 @@
 __all__ = [
     "hyper_params_factory",
     "filtering",
+    "preprocessing",
     "get_provider",
     "get_devices",
     "get_simulator",

pyriemann_qiskit/utils/preprocessing.py

@@ -0,0 +1,71 @@
+from sklearn.base import TransformerMixin
+from sklearn.preprocessing import RobustScaler
+
+
+class NdRobustScaler(TransformerMixin):
+    """Apply one robust scaler by feature.
+
+    RobustScaler of scikit-learn [1]_ is adapted to 3d inputs [2]_.
+
+    References
+    ----------
+    .. [1] \
+        https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.RobustScaler.html
+    .. [2] \
+        https://stackoverflow.com/questions/50125844/how-to-standard-scale-a-3d-matrix
+
+    Notes
+    -----
+    .. versionadded:: 0.2.0
+    """
+
+    def __init__(self):
+        self._scalers = []
+
+    """Fits one robust scaler on each feature of the training data.
+
+    Parameters
+    ----------
+    X : ndarray, shape (n_matrices, n_features, n_samples)
+        Training matrices.
+    _y : ndarray, shape (n_samples,)
+        Unused. Kept for scikit-learn compatibility.
+
+    Returns
+    -------
+    self : NdRobustScaler instance
+        The NdRobustScaler instance.
+    """
+
+    def fit(self, X, _y=None, **kwargs):
+        _, n_features, _ = X.shape
+        self._scalers = []
+        for i in range(n_features):
+            scaler = RobustScaler().fit(X[:, i, :])
+            self._scalers.append(scaler)
+        return self
+
+    """Apply the previously trained robust scalers (on scaler by feature)
+
+    Parameters
+    ----------
+    X : ndarray, shape (n_matrices, n_features, n_samples)
+        Matrices to scale.
+    _y : ndarray, shape (n_samples,)
+        Unused. Kept for scikit-learn compatibility.
+
+    Returns
+    -------
+    self : NdRobustScaler instance
+        The NdRobustScaler instance.
+    """
+
+    def transform(self, X, **kwargs):
+        _, n_features, _ = X.shape
+        if n_features != len(self._scalers):
+            raise ValueError(
+                "Input has not the same number of features as the fitted scaler"
+            )
+        for i in range(n_features):
+            X[:, i, :] = self._scalers[i].transform(X[:, i, :])
+        return X

tests/test_preprocessing.py

@@ -0,0 +1,28 @@
+import numpy as np
+from pyriemann_qiskit.utils.preprocessing import NdRobustScaler
+from sklearn.preprocessing import RobustScaler
+
+
+def test_ndrobustscaler(get_covmats):
+    n_matrices, n_features = 5, 3
+
+    X = get_covmats(n_matrices, n_features)
+
+    scaler = NdRobustScaler()
+    transformed_X = scaler.fit_transform(X)
+
+    assert transformed_X.shape == X.shape
+
+    # Check that each feature is scaled using RobustScaler
+    for i in range(n_features):
+        feature_before_scaling = X[:, i, :]
+        feature_after_scaling = transformed_X[:, i, :]
+
+        # Use RobustScaler to manually scale the feature and compare
+        manual_scaler = RobustScaler()
+        manual_scaler.fit(feature_before_scaling)
+        manual_scaled_feature = manual_scaler.transform(feature_before_scaling)
+
+        np.testing.assert_allclose(
+            feature_after_scaling, manual_scaled_feature, rtol=1e-5
+        )