Monitoring for ML

Real time graphs, comparison tables, completion estimates, and more.

Simple API

Losswise provides ML practioners with an intuitive and elegant Python API and accompanying dashboard to visualize progress within training sessions as well as across training sessions.

It's ideal for running hyperparameter searches, model comparison experiments, as well as simply tracking one-off training scripts.

              
import losswise
import time
losswise.set_api_key("my api key")
session = losswise.Session(tag='test', max_iter=10)
graph = session.graph('loss', kind='min')
for x in range(10):
    a = 1. / (0.1 + x)
    b = 1.5 / (0.1 + x)
    graph.append(x, {'train_loss': a, 'test_loss': b})
    time.sleep(1.)
session.done()
            

Interactive visualization

Points are plotted in real time as they are received.

Zoom in, select a single session, or multiple if you want to compare runs, and hover to view data points.

Summary statistics

Losswise provides real time tables showing you your latest model results, including estimates of when your models will be done training.

Sort and filter by min loss, max accuracy, or whatever you wish, to get a better understanding of how parameters affect your final results.

Smart notifications

Get notifications when your experiments succeed or when they fail.

Git tracking

Seamlessly correlate experimental results with your git branch and your git diff at the time of your experiment.

Zero overhead

Losswise uses separate threads to log data to the cloud, ensuring zero performance impact on user's applications. Training big models is slow enough at it is.

Robust

Losswise is robust against network failures. We are not in the business of crashing people's programs. If your training server loses internet access, Losswise will just print a warning to the console, and resumes logging once internet access is re-established.

Complete flexibility

Losswise believes monitoring and model tracking should be decoupled from your machine learning framework. Use Losswise to seamlessly switch between Tensorflow, Pytorch, Caffe2, MxNet, Gluon, scikit-learn, or your ML framework of choice.

"Losswise makes it easy to track the progress of a machine learning project. With Losswise and its Buildkite integration, I find I spend less time worrying about the operational complexity of training models so I can focus on other things such as improving datasets or experimenting with new model architectures."
Nicolas D. Jimenez

Nico Jimenez

Founder at Mathpix