dtaoo / dcca Goto Github PK

Hiya! Thank you for writing this up. I'm finding it interesting to play with.

I'm running into some issues that seem to be related to the "stability fix" in cca_layer.py. When we drop the components with small eigenvalues, it causes problems when we later assume these matrices are square and, also, when we try to multiply them with SigmaHat12 (which doesn't have its shape updated).

I've created a patch, and I'll submit a pull request for you to look at. The change I'm proposing seems to fix some occasional crashes, but I'm not familiar enough with the underlying math to be sure that I haven't screwed up the semantics of what the code is trying to accomplish.

Thanks,
Alex Dodge

Hi,

I am trying to run this project to explore DCCA. But model.compile() fails with the following error.

Traceback (most recent call last):
  File "dcca.py", line 73, in <module>
    model.compile(optimizer='rmsprop', loss=constant_loss) #, metrics=[mean_pred])
  File "/home/psg/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 211, in compile
    self._cache_output_metric_attributes(metrics, weighted_metrics)
  File "/home/psg/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 736, in _cache_output_metric_attributes
    output_shapes.append(list(output.shape))
  File "/home/psg/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 838, in __iter__
    raise ValueError("Cannot iterate over a shape with unknown rank.")
ValueError: Cannot iterate over a shape with unknown rank.

I am not able to identify the issue.

Python: 3.7.3
Tensorflow: 1.14.0
Keras: 2.3.1
OS: Ubuntu 18.04