Self-Adjusting Experiments
It's Easier with Glacyr.
Notebooks
Collect all data, models and glue code in a single synchronized notebook that is stored in Glacyr and always available to colleagues.
Don't Wait Twice
With self-adjusting experiments, intermediate results are cached automatically, and when modifications are made to notebooks, only the results that depend on modified code are reevaluated.
Scale with Mahler
Our open source Mahler VM is purpose built for orchestrating data-driven software running in serverless environments. Mahler can automatically split up workloads across any number of worker nodes, move running programs from one machine to another – without pausing them, and transparently recovers from faulty hardware.
In-Depth Competitions
Creating competitions on Glacyr is as easy as adding a single line of code. Leaderboards can be ranked using a custom objective function but are also ranked based on runtime and approximate energy consumption of each entry.
Keep Track of Changes
Isolate and correlate historic modifications made to experiments with current prediction performance.
Write data-driven algorithms and their corresponding experiments in collaborative notebooks.
Automatically cache intermediate results and minimize recomputation to modified lines only.
Download our iOS app to monitor progress of running experiments on the go.
Use our Mahler programming language to write concise and reproducible experiments.
Run or enter competitions among colleagues, students or teams on a shared objective.
Pause running experiments to validate their state before deciding whether to continue or not.
Install Glacyr workers on remote machines while still using the user interface from a local computer.
Accelerate training new models on CUDA powered GPUs, and run inference on other GPUs too.
Move running experiments to different hardware with no loss of progress.
Experiments on Glacyr are self-contained, deterministic and can always be reproduced from scratch.
Receive push notifications when experiments enter critical self-defined sections.
Isolate and correlate historic modifications made to experiments with their current performance.