Welcome to the MLE-Infrastructure 🔬
Experiment Logging | Parameter Searches | Experiment Launch | Experiment Protocol | Experiment Manager |
mle-logging |
mle-hyperopt |
mle-scheduler |
mle-monitor |
mle-toolbox |
The MLE-Infrastructure provides a reproducible workflow for distributed Machine Learning experimentation (MLE) with minimal engineering overhead. The core consists of 5 packages:
mle-logging
: Experiment logging with easy multi-seed and configuration aggregation.mle-hyperopt
: Hyperparameter Optimization with config export, refinement & reloading.mle-monitor
: Monitor cluster/cloud VM resource utilization & protocol experiments.mle-scheduler
: Schedule & monitor jobs on Slurm, GridEngine clusters & GCP VMs.mle-toolbox
: Glues everything together to manage & post-process experiments.
Note I: A template repository of an infrastructure-based project can be found in the mle-project
. You can inspect your experiment stack in an interactive web UI: mle-laboratory
.
Note II: mle-logging
, mle-hyperopt
, mle-monitor
and mle-scheduler
are standalone packages and can be used independently of the utilities provided by the mle-toolbox
.