Scikit-learn transformer example

Here we demonstrate a simple application of scattering as a transformer

Preliminaries

Import the relevant classes and functions from sciki-learn.

from sklearn.pipeline import Pipeline
from sklearn.model_selection import train_test_split
from sklearn import datasets
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import StandardScaler

Import the scikit-learn Scattering2D frontend.

from kymatio.sklearn import Scattering2D

Preparing the data

First, we load the dataset. In this case, it’s the UCI ML digits dataset included with scikit-learn, consisting of 8×8 images of handwritten digits from one to ten.

digits = datasets.load_digits()

We then extract the images, reshape them to an array of size (n_features, n_samples) needed for processing in a scikit-learn pipeline.

images = digits.images
images = images.reshape((images.shape[0], -1))

We then split the images (and their labels) into a train and a test set.

x_train, x_test, y_train, y_test = train_test_split(images, digits.target,
                                                    test_size=0.5, shuffle=False)

Training and testing the model

Create a Scattering2D object, which implements a scikit-learn Transformer. We set the input shape to match that of the the images (8×8) and the averaging scale is set to J = 1, which means that the local invariance is 2 ** 1 = 1.

S = Scattering2D(shape=(8, 8), J=1)

We then plug this into a scikit-learn pipeline which takes the scattering features, scales them, then provides them to a LogisticRegression classifier.

classifier = LogisticRegression(max_iter=150)
estimators = [('scatter', S), ('scaler', StandardScaler()), ('clf', classifier)]
pipeline = Pipeline(estimators)

Given the pipeline, we train it on (x_train, y_train) using pipelien.fit.

pipeline.fit(x_train, y_train)

Pipeline(steps=[('scatter',
                 Scattering2D(J=1,
                              backend=<class 'kymatio.scattering2d.backend.numpy_backend.NumpyBackend2D'>,
                              shape=(8, 8))),
                ('scaler', StandardScaler()),
                ('clf', LogisticRegression(max_iter=150))])

Finally, we calculate the predicted labels on the test data and output the classification accuracy.

y_pred = pipeline.predict(x_test)

print('Accuracy:', accuracy_score(y_test, y_pred))

Accuracy: 0.9755283648498332