Training the tICs instead of the features #6
Description
I generated 4 tICs from the 4 dihedral features presented in the attached notebook. Now I want to train the tICs to see which tIC is more dominant over the other. And in this regard I want to generate a plot similar like SVM co-efficients vs Feature index mentioned in this notebook https://github.com/msultan/SML_CV/blob/master/alanine_example/01-svm_example.ipynb .
First question:
I tried something like that just substituting the original script a bit to train the tICA features
X=np.vstack(plot_feat)
train_X=np.vstack(tica_features)
y=np.concatenate([np.zeros(len(plot_feat[0])),
np.ones(len(plot_feat[0]))])
if train:
clf.fit(train_X,y)
train_X.sum(axis=1)[300:].std()
What is the meaning of this line train_X.sum(axis=1)[300:].std() and the output 2.2081068901447987
The I plotted the tIC0 vs SVM co-efficient using the following script and got an output like following
plot([0,1,2,3],clf.coef_[0],color=sns.color_palette("colorblind")[0],marker='o',label=r'$\alpha_L$ vs $\beta$')
xticks([0,1,2,3],[r'tic0',r'tic1',r'tic2',r'tic3'],size=20)
xlabel("tICs Index")
ylabel(r'$SVM_{cv}$ Coefficients')
legend()
Simple question: How do I get the exact value of SVM coefficient for each tICs. It is a bit hard to see from the plot. Any script?