Feature/pca 2d projection #75
Conversation
scikitplot/decomposition.py
Outdated
| title_fontsize="large", text_fontsize="medium"): | ||
| figsize=None, cmap='Spectral', | ||
| title_fontsize="large", text_fontsize="medium", | ||
| biplot=False, feature_labels=None): |
There was a problem hiding this comment.
These new arguments should be before ax
scikitplot/decomposition.py
Outdated
|
|
||
| feature_labels (array-like, shape (n_classes), optional): List of labels | ||
| that represent each feature of X. It's index position must also be | ||
| relative to the features. If none is given, then labels will be |
scikitplot/decomposition.py
Outdated
| biplots. If false, the biplots are not generated. | ||
|
|
||
| feature_labels (array-like, shape (n_classes), optional): List of labels | ||
| that represent each feature of X. It's index position must also be |
scikitplot/decomposition.py
Outdated
| @@ -156,14 +166,36 @@ def plot_pca_2d_projection(clf, X, y, title='PCA 2-D Projection', ax=None, | |||
| classes = np.unique(np.array(y)) | |||
|
|
|||
| colors = plt.cm.get_cmap(cmap)(np.linspace(0, 1, len(classes))) | |||
|
|
|||
|
|
|||
| vectors=np.transpose(clf.components_[0:2, :]) | |||
There was a problem hiding this comment.
Style: Add spaces around equals sign.
Also, not sure about this but can 0:2 be written as simply :2?
There was a problem hiding this comment.
Sry, Bad habit on indexing, I'll fix that.
scikitplot/decomposition.py
Outdated
| vectors=np.transpose(clf.components_[0:2, :]) | ||
|
|
||
| xs = transformed_X[:,0] | ||
| ys = transformed_X[:,1] |
There was a problem hiding this comment.
vectors, xs, and ys should be evaluated within the if biplot: block since they're only used there.
scikitplot/decomposition.py
Outdated
| xs = transformed_X[:,0] | ||
| ys = transformed_X[:,1] | ||
|
|
||
|
|
scikitplot/decomposition.py
Outdated
| if biplot: | ||
| for i in range(vectors.shape[0]): | ||
| ax.annotate("", xy=(vectors[i,0]*max(xs), | ||
| vectors[i,1]*max(ys)), xycoords='data', |
scikitplot/decomposition.py
Outdated
|
|
||
| if feature_labels is None: | ||
| plt.text(vectors[i,0]*max(xs)*1.05, vectors[i,1]*max(ys)*1.05, | ||
| "Variable"+str(i), color='b', fontsize=text_fontsize) |
There was a problem hiding this comment.
Style: Space after comma, space around operators
scikitplot/decomposition.py
Outdated
| "Variable"+str(i), color='b', fontsize=text_fontsize) | ||
| else: | ||
| plt.text(vectors[i,0]*max(xs)*1.05, vectors[i,1]*max(ys)*1.05, | ||
| feature_labels[i], color='b', fontsize=text_fontsize) |
There was a problem hiding this comment.
I think the if-else branch here should only set a single variable feature_label since the rest of the calculations are duplicated.
There was a problem hiding this comment.
Actually this is cleaner
plt.text(vectors[i,0]*max(xs)*1.05, vectors[i,1]*max(ys)*1.05,
feature_labels[i] if feature_labels else "Variable" + str(i), color='b', fontsize=text_fontsize)
scikitplot/decomposition.py
Outdated
| ax.legend(loc='best', shadow=False, scatterpoints=1, | ||
| fontsize=text_fontsize) | ||
| ax.set_xlabel('First Principal Component', fontsize=text_fontsize) | ||
| ax.set_ylabel('Second Principal Component', fontsize=text_fontsize) | ||
| ax.tick_params(labelsize=text_fontsize) | ||
|
|
||
| return ax | ||
|
|
There was a problem hiding this comment.
Should only have single newline at end of file
|
First time I used pylint, solid Haha. Just let me know if I missed anything. |
scikitplot/decomposition.py
Outdated
| xs = transformed_X[:, 0] | ||
| ys = transformed_X[:, 1] | ||
| for i in range(vectors.shape[0]): | ||
| ax.annotate("", xy=(vectors[i, 0]*max(xs), vectors[i, 1] * max(ys)), |
There was a problem hiding this comment.
Maybe you could do this calculation outside the for loop since it's duplicated e.g.
vectors = np.transpose(clf.components_[:2, :])
xs = transformed_X[:, 0]
ys = transformed_X[:, 1]
vectors_scaled = vectors * [max(xs), max(ys)]
Then inside the for loop:
ax.annotate("", xy=vectors_scaled[i], ...
Btw I'm not sure what vectors_scaled actually is so you can probably name it something else.
Also, haven't tested this code so might have made a mistake somewhere. Idea is to take advantage of numpy's matrix multiplication instead of multiplying inside a Python for loop.
There was a problem hiding this comment.
Yea, I can work on a test.
You are absolutely correct, I should have considered matrix multiplication.
vectors_scaled seems like an appropriate enough name. Its simply a multiplier to scale the biplot vectors so that they may be better visualized in the pca plane.
scikitplot/decomposition.py
Outdated
| xycoords='data', xytext=(0, 0), textcoords='data', | ||
| arrowprops={'arrowstyle': '-|>', 'ec': 'r'}) | ||
|
|
||
| plt.text(vectors[i, 0] * max(xs) * 1.05, vectors[i, 1] * max(ys) * 1.05, |
There was a problem hiding this comment.
Then here you could write vectors_scaled[i, 0] * 1.05, vectors_scaled[i, 1] * 1.05
|
Merged. Thanks a lot 👍 |
addressed enhancement on #21.
Also updated the pca_2d_projection notebook example to show biplots.