Recently, I finally begin to do the practice of deep learning network. It’s easy to tell the technologies e.g. BP, dropout, BN and pooling. However, sometimes it can be really difficult to configure and execute the network on the machine.
Previously I didn’t have any desktop with GPU and the experience of using Uni’s lab is a bad experience. They assume you know exactly what you are working on. But in the real world, we have to learn from experiments. Even it’s a bad experiments e.g. breaking the system 😛
In order to have a good experience and the full control of root permission, I recently got one desktop with RTX2080TI, one of the best gaming GPUs in the market. The first step is to install win10 and ubuntu18 on it. Those are the easy part and you can find many tutorials online.
The difficult part is to install the caffe and anaconda3.
I will skip all the common steps and focus on the wired issues I have found:
Caffe should be compiled on the local with make. Although for ubuntu18 we can directly download the caffe with apt install. But if you need to change any configuration of it, you have to compile it manually. That means we need to download the caffe from github and run make
If we want to use caffe with anaconda3,we need to modify the makefile.config. The good thing is it has an existing template of how to configure anaconda3.
Don’t put the path of anaconda/lib into the entire system. It may cause issue for the system booting.
If you need the path of anaconda/lib, put it into the .bashrc.
If you want to use import caffe, you need to manually compile the make pycaffe