htuneml 0.0.6

Monitor machine learning experiments

# HTUNEML - machine learning experiments monitoring and tuning

Quickstart: pip install htuneml. See the “Installing” section for more details.

Project links:

• PyPI

• Source code

• Issue tracker

Examples

See the examples/ directory in the repository root for usage examples:

• Tensorflow Mnist

• Keras Mnist

• Pytorch Mnist

Requirements

To use all of the functionality of the library, you should have:

• Python 2.6, 2.7, or 3.3+ (required)

• PyAudio 0.2.11+ (required only if you need to use microphone input, Microphone)

Quick start

Register on website http://registru.ml, copy the api_key:

import htuneml as ht

job = Job('api_key')

@job.monitor
def train(par1=2,par2=2):
    for i in range(par1):
        #do training here
        job.log({'loss':i*4,'ep':i})

job.setName('l2')
#job.debug()# uncomment and no experiment will be created and no logs sent
train(10, 2)

This will print out something like the following:

make experiment
got key experimnet 5c5c8eaacbcfb9146641367a

Also it is possible to sent the parameters from the web app. First on gpu/cpu set the lisener:

import htuneml as ht

job = Job('api_key')

def train(par1=2,par2=2):
    for i in range(par1):
        #do training here
        job.log({'loss':i*4,'ep':i})

job.sentParams(train)#sent the parameters list to the app
job.waitTask(train)#wait for parameters from app