Load and save input similarities

[dkobak]

I moved your unused code to load/save input similarities into run(). There is an additional input parameter that specifies whether Ps should be loaded, saved, or nothing. I added it to the all three wrappers (but only tested in Python).

What is loaded/saved are Ps after symmetrizing, so that's something I needed to change compared to your code (because the size of val_P/col_P changes after symmetrizing).

Also, if theta=0 then what is loaded/saved is the full P array.

[linqiaozhi]

Hey @dkobak, this looks great! The R test script runs nicely, but running the Matlab test script examples/test.m gives an error:

>> cluster_firstphase = fast_tsne(X_clusters,opts);
Undefined function or variable 'load_affinities'.

Error in fast_tsne>write_data (line 299)
    fwrite(h, load_affinities, 'integer*4');

Error in fast_tsne (line 250)
    write_data('temp/data.dat',X, no_dims, theta, perplexity, max_iter, ...

It looks like load_affinities is not passed to the function write_data. If that's the bug, can you please fix it, commit, and I can go ahead and merge?

[dkobak]

Oops. Done.

[linqiaozhi]

Great, thanks.

The loading actually still does not work in Matlab, because of the following lines.

https://github.com/dkobak/FIt-SNE/blob/master/fast_tsne.m#L113-L115

If you delete these lines, it should work.

[dkobak]

True, I did not notice those. Deleted.

[linqiaozhi]

Merged, thanks so much, @dkobak!!