Features¶

Here we showcase some of the features of Qanat.

Experiment run tracking¶

../_images/experiment_basic.gif

Automatic creation of experiment results directory
Automatic logging of experiment parameters, git commit hash, stdout, stderr, etc.
History of experiment runs with date, duration, description, tags, etc.

Ease of experiment running¶

../_images/experiment_htcondor_container.gif

Launch experiments with a grid over parameters easily
Facilitate run on job submission systems: HTCondor and Slurm
Facilitate run on local machine of several parameters in parallel
Facilitate run inside a container (Docker, Singularity)
Track experiment progress

Dataset handling¶

Track which dataset used for which experiment
Automatically bind mount dataset directory inside container

Analysis of experiments results¶

Explore run of experiments by searching over details of experiment runs
Analysis script after runs are finished can be formalised as actions on the experiment results

Documents handling¶

Track documents relative to the project with tags, description, etc.
Add experiment run dependencies on documents

Reproducibility¶

Track git commit hash of the experiment code
Reproduce experiment run from previous run in the same environment

Templates¶

Many templates for different types of projects are available

Simple python project
Simple Matlab project
Simple Julia project
Multivariate Statistics project
Deep Learning project
etc.

Roadmap¶

[ ] Add support for Docker
[ ] Add experiment to experiment dependencies
[ ] Add hosts functionality to track different running machines
[ ] Update document, experiments, dataset as an edit of YAML file
[ ] Edit database directly in case of problems