Added Documentation regarding issue #1878

[kislayabhi]

Added 'pca_notebook.ipynb' named python notebook in
doc/ipython-notebooks/pca
Implemented PCA on toy data for 2d to 1d and 3d to 2d projection.
Implemented Eigenfaces for data compression and face recognition
using att_face dataset.

[vigsterkr]

great! but please make sure that the output of the notebook is always clear, we don't want to store the generated output of the notebook in the repository.

[karlnapf]

Thanks!

One thing. Could you always post an nbviewer link to the current version of the notebook in the PR (with output)
The PR itself should not contain any output, so clear before you commit.

[karlnapf]

Data should be added to shogun-data (if open) and then the notebook should reference that.
Downloading manually is not a choice

[kislayabhi]

@karlnapf , yeah!! I will amend those.

[kislayabhi]

nbviewer link:
http://nbviewer.ipython.org/gist/kislayabhi/9431770

[karlnapf]

We have a template for ipython notebooks, pls use the style suggested in there regarding headers, title, author list etc.

[karlnapf]

I really like where this is going, but there are currently many issues that need to be addressed.
Notebooks in general needs loads of love to look nice and convincing.
Again, pls have a look into other ones such as the GP to get some idea where we want to go with those notebooks

[karlnapf]

And thanks for the hard work :)

[vigsterkr]

just one more thing: .ipynb_checkpoints directory or anything under it should not be added to repository

[kislayabhi]

@karlnapf , @vigsterkr : I will get it done! Thanks for evaluating it. About the data, I sent a PR yesterday to the shogun-data with my dataset updated. Did I do something wrong there?

Again thanks. You have given me a direction to work on!! loving it :)

[karlnapf]

Great! Let us know when you have an updated version. You should put it under the same gist/nbviewer link.

[kislayabhi]

updated :) (http://nbviewer.ipython.org/gist/kislayabhi/9431770)
Phew, it took me some time.
Sorry for the delay

[karlnapf]

"Here the best basis vectors are defined as"
Its not the best basis vectors, just the basis vectors

[karlnapf]

"The optimal bias c is given"
Same here, bias is not optimal in the squared error, its just some bias

[karlnapf]

"To find the best basis vectors B " Here we are talking about the best, so thats fine :)

[karlnapf]

( i.e the basis vectors are mutually orthogonal and of unit length )

no whitespace after ( and before )

[karlnapf]

EGD -- should be EVD I guess?

[karlnapf]

"from matplotlib import pyplot" You can assume that our notebook server is started with
ipython notebook --pylab inline so there is not need to import anything. Also, all functions are loaded to the std namespace, so you can just do plot(...) instead of pylab.plot(...)

[karlnapf]

Step 4: Calculate the eigenvectors and eigenvalues of the covariance matrix
Could you initialise PCA with explicitly using the constructors options and commenting on them? (not in the source code, but in notebook)

In fact, before you start with the PCA code, could you add a litlle paragraph on the CPCA class itself, quickly comment on how its initialised (see above, constructor parameters), and what type of data it accepts, what type of preprocessor it is, that it is a preprocessor in Shogun that is used in a certain way, etc? This is all Shogun specific, but its important to also document those points

[karlnapf]

And in the above, also add link to the Shogun class list

[karlnapf]

"Steps 5" there is an "s" too much

[karlnapf]

"<matplotlib.text.Text at 0x3764bd0>"

Could you hide those type of outputs via _=plot(...)

[karlnapf]

"Step6: Deriving the new data set"
What about "Projecting the data to its principal conponents", also a whitespace missing

[karlnapf]

matrices (with D feature dimensions) supplied via apply_to_feature_matrix methods.This tranformation
#outputs the M-Dimensional approximation of all these input vectors and matrices (where M<=min(D,N)).

please dont do such things in comments, but rather in the notebook directly

[karlnapf]

"preprocessor(from next example)" whitespace

[karlnapf]

`````` ax.set_zlim(-30,30)please do_=ax.set_zlim(-30,30)``` to hide any python output for such plotting calls

[karlnapf]

#E is automagically filled by setting target dimension = M. This is different from the 2d data example
#where we implemented this step manually.

Please write this in the notebook rather than in a comment

[karlnapf]

"PCA Performance"
I just saw this, maybe this should be moved a bit to the top. I commented that its missing earlier

[karlnapf]

Two-dimensional p*q grayscale should be "pq" (in latex), so no *

[karlnapf]

"Step 1:Get" whitespace

[karlnapf]

np.array etc can be array when you start the notebook with --pylab inline

[karlnapf]

Euclidean distance - please add a link to the class documentation
http://www.shogun-toolbox.org/doc/en/latest/classshogun_1_1CEuclideanDistance.html

In fact, please do this for all classes you mention

[karlnapf]

Wow, I have to say I am really impressed by this notebook! :)
It is very nicely written, and the examples are just amazing.
I hope you dont mind all my comments, it got so much better since the first iteration.
I think once my current issues are resolved, I will iterate/read once more and then we can merge this, and tell the world how great it is. Please add a comment in the NEWS file
-new PCA notebook by [yourname]
Great work!

[karlnapf]

BTW I agree on your last point with the np.dot.
Also I hope your exams went well!

[kislayabhi]

Please refer to the following pull request.
pca ipython notebook revised 3rd iteration #2028
I somehow messed with my earlier forked repo :(

I have updated the nbviewer file