You have to a system with pyhton3
and pip
. I recommend install virtualenv
to set an environment on
your project folder, then, with command install -r requirements.txt
, all dependencies will be installed.
For some graph pyhton3-tk
is required so I need to install it with apt-get install pyhton3-tk
You need to download the dataset (in .mat
format) from here and copy
all files in svhn
folder.
This code is composed by three python files:
load.py
reshape and format dataset to using in training and testing modedp.py
describe network configuration and compute performancemain.py
build the network and run training or test phase
be careful
If you run this code with combination of training and extra dataset you need a more then 16gb RAM and 8 core CPU to complete training in about 2 hours
You can visualize your network trough your browser. You have to activate your virtualenv and run tensorboard command
~/$ source bin/activate
(ProjectName) ~/$ python -m tensorflow.tensorboard --logdir=/path/to/board/folder