This is a Nvidia-docker image contains Machine learning tools I used. Build on ubuntu16.04 , CUDA 8.0. It will fit any Ubuntu computers with Nvidia-driver installed.
To use this docker, you must have the latest docker and nvidia-docker and driver installed, please check out this.
Installed ML packages:
- cuda 8.0, cudnn 5110
- opencv 3.2
- numpy scipy
Updating via apt and pip
If you found the updating via apt/pip is too slow, you have to reconfigure the source in
/etc/apt/sources.list or ~/.pip/pip.conf
When you use caffe, you many encounter (error == cudaSuccess (8 vs. 0) invalid device function) , though theano, tensorflow, keras could use GPU with no problem. I have no idea what causes this error but rebuild caffe could help:
pip uninstall protobuf cd ~/downloads/caffe/build rm -rf ./* cmake ../ make -j 4 make all make install pip install protobuf
cd ~/downloads/caffe/ ./examples/mnist/train_lenet.sh