losswise 4.0

Official Losswise library for Python

This is the official Losswise Python library. This library allows for server-side integration of Losswise.

Installation

The library can be installed using pip:

pip install losswise

Getting Started

First create an account on the Losswise website (https://losswise.com). This will automatically generate a unique API key.

Typical usage usually looks like this:

import random
import losswise

# replace with your own api key
losswise.set_api_key('your_api_key')

# replace with a string that identifies your model
session = losswise.Session(tag='my_dilated_convnet', max_iter=10, data={'num_params': 10000000})

# create empty graph for loss, keep track of minima here hence kind='min'
graph = session.graph(title='loss', kind='min')

# track artificial loss over time
for x in xrange(10):
    train_loss = 1. / (0.1 + x + 0.1 * random.random())
    test_loss = 1.5 / (0.1 + x + 0.2 * random.random())
    graph.append(x, {'train_loss': train_loss, 'test_loss': test_loss})

# mark session as complete
session.done()

You can then view the visualization results on your dashboard.