David Kelley

I'm also interested in this functionality.

Curious if anyone has had any luck implementing the 1D version. There's no depthwise_conv1d, and the standard playbook to expand_dims and use depthwise_conv2d returns error "Current implementation only supports equal...

HashSolo is meant for working with hashing data. Are you simply attempting to call doublets? We haven’t worked with a data set with only 800 cells before. Solo benefits from...

The output file is_doublets.npy contains binary doublet calls. You can see that and all of the other output files in the lines of code after this one: https://github.com/calico/solo/blob/master/solo/solo.py#L394

If you think your data has 800 cells, but is_doublets.npy has 10k values, then it's likely the input data isn't formatted according to Solo's assumptions. Could you send me more...

Are you saying that you don’t understand how to read the vector stored in is_doublets.npy? Or are you saying that you don’t understand how your 12,013 cells was filtered down...

OK, you'll want to read the is_doublets.npy file using a command like the following in a python terminal, notebook, script. cell_doublets = np.load('is_doublets.npy') cell_doublets will contain a numpy array with...

Are you referring to this tutorial? https://github.com/calico/basenji/blob/master/manuscripts/akita/tutorial.ipynb I wasn't able to reproduce the error using the same tensorflow version. Are you sure that you didn't change anything before running? Maybe...

Great. No powers of 2 is not a requirement. But whatever you decide, you need to match the max pooling widths in the convolution tower. So if you wanted to...

Which script are you referring to for `-l`? I'm guessing `akita_data.py`. Are you saying that you're able to generate TFRecords, but your training crashes? If so, what did you set...