Dimension problem

[lvdmaaten]

Dimension problem

[lvdmaaten]

I have never used h5py, does it return a NumPy array?

The most likely explanations would be your process not having permissions to create a temporary directory here or a temporary file here. This will be difficult to debug for me without a repro.

[dcharua]

Thank you for the quick response, It does returns a numpy array if I run data.shape I get (2462, 3, 32, 32).
so the error is here:

File "bhtsne.py", line 112, in init_bh_tsne
cov_x = np.dot(np.transpose(samples), samples)
ValueError: shapes (32,32,3,2462) and (2462,3,32,32) not aligned: 2462 (dim 3) != 32 (dim 2)

Something about the dimensions doesn't add up when doing the .dot between the transpose and the original matrix, maybe since is a big array with the whole dataset.

How can I import my dataset? I can try to read directly from the folders into a np.array but in what format does it needs to be? an array for each image? or a 4d array with all images? and does it need the labels?

Thank you very much

[lvdmaaten]

Your dataset should be a matrix of size NxD, where N is the number of examples and D is the number of dimensions. Please reshape your data matrix to size 2462x(3x32x32) before feeding it to t-SNE.