ESNRegressor, GridSearchCV, Pipeline

In [84]:

"""
=========================================================
scikit-learn Pipeline and GridSearch with ESNRegressor 
=========================================================
This example shows a minimal example of Echo State Network
applied to a regression problem using a sklearn Pipeline 
and a GridSearch with cross-validation.
"""
import numpy as np
from sklearn.metrics import r2_score
from sklearn.model_selection import train_test_split, GridSearchCV, TimeSeriesSplit
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

from echoes import ESNRegressor
from echoes.plotting import plot_predicted_ts

# Prepare synthetic data 
x = np.linspace(0, 30*np.pi, 1000).reshape(-1,1)
inputs = 100 * np.sin(x) + np.random.normal(scale=.2, size=x.shape)
outputs = np.cos(x) + np.random.normal(scale=.2, size=x.shape)
X_train, X_test, y_train, y_test = train_test_split(inputs, outputs, test_size=.3, shuffle=False)

pipe = Pipeline(
    [
        ("scaling", StandardScaler()),  # Try commenting out this line to see the effect
        ("esn", ESNRegressor(n_transient=100))
    ]
)

# Pass the parameter to sklearn pipeline with prefix esn__
param_grid={
    "esn__n_reservoir": [20, 50, 100, 150],
    "esn__leak_rate": [0.5, 0.8, 0.9, 0.95],
    "esn__spectral_radius": [0.85, 0.9, 0.95, 0.99],
}

grid = GridSearchCV(
    pipe,
    param_grid=param_grid,
    n_jobs=5,
    cv=TimeSeriesSplit() # Use time series folds generator  
)
grid.fit(X_train, y_train)

""" ========================================================= scikit-learn Pipeline and GridSearch with ESNRegressor ========================================================= This example shows a minimal example of Echo State Network applied to a regression problem using a sklearn Pipeline and a GridSearch with cross-validation. """ import numpy as np from sklearn.metrics import r2_score from sklearn.model_selection import train_test_split, GridSearchCV, TimeSeriesSplit from sklearn.pipeline import Pipeline from sklearn.preprocessing import StandardScaler from echoes import ESNRegressor from echoes.plotting import plot_predicted_ts # Prepare synthetic data x = np.linspace(0, 30*np.pi, 1000).reshape(-1,1) inputs = 100 * np.sin(x) + np.random.normal(scale=.2, size=x.shape) outputs = np.cos(x) + np.random.normal(scale=.2, size=x.shape) X_train, X_test, y_train, y_test = train_test_split(inputs, outputs, test_size=.3, shuffle=False) pipe = Pipeline( [ ("scaling", StandardScaler()), # Try commenting out this line to see the effect ("esn", ESNRegressor(n_transient=100)) ] ) # Pass the parameter to sklearn pipeline with prefix esn__ param_grid={ "esn__n_reservoir": [20, 50, 100, 150], "esn__leak_rate": [0.5, 0.8, 0.9, 0.95], "esn__spectral_radius": [0.85, 0.9, 0.95, 0.99], } grid = GridSearchCV( pipe, param_grid=param_grid, n_jobs=5, cv=TimeSeriesSplit() # Use time series folds generator ) grid.fit(X_train, y_train)

Out[84]:

GridSearchCV(cv=TimeSeriesSplit(gap=0, max_train_size=None, n_splits=5, test_size=None),
             estimator=Pipeline(steps=[('scaling', StandardScaler()),
                                       ('esn', ESNRegressor(n_transient=100))]),
             n_jobs=5,
             param_grid={'esn__leak_rate': [0.5, 0.8, 0.9, 0.95],
                         'esn__n_reservoir': [20, 50, 100, 150],
                         'esn__spectral_radius': [0.85, 0.9, 0.95, 0.99]})

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

In [85]:

y_pred = grid.best_estimator_.predict(X_test)  # Get prediction for plotting
print("test r2 score: ", round(grid.score(X_test, y_test), 3))
plot_predicted_ts(y_test, y_pred, figsize=(15, 5), start=100);

y_pred = grid.best_estimator_.predict(X_test) # Get prediction for plotting print("test r2 score: ", round(grid.score(X_test, y_test), 3)) plot_predicted_ts(y_test, y_pred, figsize=(15, 5), start=100);

test r2 score:  0.916

In [ ]: