tensor-observer 0.0.4

A little api library for a TensorObserver server

tensor-observer

Simple PHP Server to keep track of current machine learning trainings.

1. Install a little api library

    pip install tensor-observer

2. push the following files to your server and maintain the folder structure

• index.php
• php/*
• assets/*
• and create a "data" folder

2. create an object of the API

from tensor-observer import TensorObserver

to_api = TensorObserver('http://localhost:8080')

4. push your scalars during training by

to_api.scalar(self, scalar, run, tag, step, wall_time=None)

This method performs a http post request to save a scalar value from the training process.

Args:

• scalar: number that should be saved
• run: run identifier
• tag: tag of the scalar
• step: step id
• wall_time: if not specified, current time is taken

5. wrap your script in a try catch block to push exceptions

to_api.exception(self, e, run, wall_time=None)

This method performs a http post request to notify that an exception happend.

Args:

• e: Exception object
• run: run identifier
• wall_time: if not specified, current time is taken

6. send a end signal at the end to mark a run as terminated

to_api.end_signal(self, run, wall_time=None)

This method performs a http post request to notify that the run terminated

Args:

• run: run identifier
• wall_time: if not specified, current time is taken

7. have fun!