Skip to content

Make the CNN training easier

Would be very nice to have a script called train.py that receives, as argument, a python script with the network configuration.

Follow bellow a possible docopt description

Train a Neural network using bob.learn.tensorflow

Usage:
  train.py [--batch-size=<arg> --validation-batch-size=<arg> --iterations=<arg> --validation-interval=<arg> --output-dir=<arg> --use-gpu --prefetch ] <configuration>
  train.py -h | --help
Options:
  -h --help     Show this screen.
  --batch-size=<arg>   [default: 32]
  --validation-batch-size=<arg>    [default: 128]
  --iterations=<arg>   [default: 1000]
  --validation-interval=<arg>   [default: 100]
  --output-dir=<arg>    [default: ./logs/]

The python script could be defined as the following

datashuffler = ...
architecture = ....
learning_rate = ...
solver = ....
loss=...
trainer=...
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information