Anders Boesen Lindbo Larsen

The CPU implementation of the convolutions is horribly slow and should not be used for training. I have never used them to train anything above MNIST-size data. :)

Hi Sergey! Just return the probabilities: ``` def fprop(self, x): return self._softmax(x) ``` I think you should also change the behavior of `y_shape()` to match the new array size. You...

Hi, I believe the scaling term is incorporated when I [precompute the Gram matrix](https://github.com/andersbll/neural_artistic_style/blob/master/style_network.py#L108)

Sorry about the late reply! It seems that CUDArray is installed incorrectly in your Python environment. Maybe you have forgotten to run `python setup install` for CUDArray.

Hey, I'm occupied at the moment. I'll get back to you in a few days.

Hey, thank you for bringing up this issue. There is definitely something fishy going on. I played around with the scale of the noise in the initial image and it...

Try `make clean` and `python setup.py clean` before building.

conv_bc01.c gets generated by the call to [cythonize()](https://github.com/andersbll/cudarray/blob/master/setup.py#L78). Regarding your initial problem: You might need to call `sudo python setup.py install` because you are trying to install to system directories....

You need to 1) implement your CUDA functionality, 2) wrap the functionality in Cython and 3) expose it to Python using `cudarray.ndarray` objects.

Sorry, can't help you with that.