Problems with data scaling in sklearn.cross_decomposition.PLSRegression #6002
Description
I am trying to fit some spectral data using PLS, and am having difficulties with the module.
Essentially, when I use the default value of scale=False, I get a prediction, BUT all my predictions are scaled, and I just cannot figure out how to convert back to my original data space. The same is true for the example code. As you see, the scaled prediction is just off. I assumed that I should be able to revert back to the "unscaled" data using pls_scaled.y_mean_ and pls_scaled.y_std_, but that doesn't seem to work for me. Any suggestions would be highly appreciated.
Y = np.array([[0.1, -0.2], [0.9, 1.1], [6.2, 5.9], [11.9, 12.3]])
X = np.array([[0., 0., 1.], [1.,0.,0.], [2.,2.,2.], [2.,5.,4.]])
pls_not_scaled=PLSRegression(n_components=2, scale=False)
pls_scaled=PLSRegression(n_components=2, scale=True)
pls_not_scaled.fit(X,Y)
Out[]: PLSRegression(copy=True, max_iter=500, n_components=2, scale=False, tol=1e-06)
pls_scaled.fit(X,Y)
Out[]: PLSRegression(copy=True, max_iter=500, n_components=2, scale=True, tol=1e-06)
Out[]: PLSRegression(copy=True, max_iter=500, n_components=2, scale=False, tol=1e-06)
Y_not_scaled_pred = pls_not_scaled.predict(X)
Y_scaled_pred = pls_scaled.predict(X)
print(Y)
[[ 0.1 -0.2]
[ 0.9 1.1]
[ 6.2 5.9]
[ 11.9 12.3]]
print(Y_not_scaled_pred)
[[ 0.11029323 -0.09323388]
[ 0.86825958 0.77077371]
[ 6.24049125 6.31999396]
[ 11.88095594 12.1024662 ]]
print(Y_scaled_pred)
[[ 1.52568016 1.47577156]
[ 2.43367318 2.3584298 ]
[ 6.25638942 6.26638408]
[ 8.88425724 8.99941456]]
For those with a more visual bend:
plt.grid('on')
plt.scatter(Y[:, 0], Y_not_scaled_pred[:, 0], color='blue')
plt.scatter(Y[:, 0], Y_scaled_pred[:, 0], color='red')
plt.xlabel('Predicted')
plt.ylabel('Measured')
plt.legend(['not_scaled','scaled'],'lower right')
