-
Notifications
You must be signed in to change notification settings - Fork 20
cross validation is slow #90
Comments
Thanks for making this issue. Is it possible that your ResNet early layer is larger than usual? It looks like extracting activations is fairly quick (2min21s). At the next step in the process, it should test the alignment of every layer to the public neural data in a cross-validated way. For this, it uses a linear regression metric so I'm wondering if the size of the activations matrix is very large and slowing things down. |
That doesn't look so big. Do you know how many units are in each layer with the pixel input size you're using? If you exclude the first layer from the layer selection, does it run much faster? The other thing could be a slow CPU or small memory size requiring activations to be swapped in and out. What's your hardware like? |
Input image is 64x64x3. NVIDIA-SMI 515.43.04 |
could you also share information on RAM / memory size? Your processors look great and like they should run through this much more quickly. |
Okay so you also have more than enough memory available. I am not sure what is causing this. You might be able to use |
Hi
I am working with public benchmarks for IT and V4 layers. My goal is to get scores for a modified 2-block resnet architecture for each layer on the benchmarks to select layers to run on private benchmarks.
It is taking me around 7 hours for encoder.conv1 layer out of which cross validation takes ~5-6 hours and similar for other layers as well.
Attaching screenshot for reference.
The text was updated successfully, but these errors were encountered: