`mle-laboratory` 👉 StreamLit app

I dream of having a WebUI through which can generate/schedule new experiments, filter experiments by project_name and generate/download reports of the results. All just by clicking a couple of times. At the same time this WebUI can act as an extended cluster 'cockpit' and should be password protected. The web page can be served from one of the cognition/standard nodes (with internet access) and will have a small footprint.

I believe that StreamLit can provide all of these functionalities, allows for fast/easy development and can also directly execute bash commands (e.g. run-experiment or report-experiment. Furthermore, downloading files via a click is also easy.

It may make sense to keep this in a separate repository (aka mle-laboratory) which depends on mle-toolbox and to create a separate organization. Currently, I have implemented a small (but ugly) prototype in Django (see below). I would start by reimplementing the protocol overview -- potentially refactoring and reusing the code used in monitor-cluster to get the data from the local database. Here is a list of steps:

[ ] Setup streamlit locally and implement basic aesthetics.
[ ] Add protocol summary overview with all experiments, status, etc.
[ ] Add details to experiments (clickable config details)
[ ] Add download of configs/generated reports via click
[ ] Add resource overview of cluster (SGE, HPC, GCP)
[ ] Add clickable generation of report by calling report-experiment
[ ] Add clickable execution of new experiment
- [ ] Generate new job config from template (set search params, etc.)
- [ ] Give base directory from which to execute
- [ ] Execute from specified directory via run-experiment
[ ] Add password protection
[ ] Host on node/remote server

mle-infrastructure / mle-toolbox

`mle-laboratory` 👉 StreamLit app #40