Getting Started

Command Line Interface (CLI) & Jupyter Notebooks

CLI allows users to create new projects leveraging the structure and abstractions of Sapsan providing a unified interface of interaction with the experiments. In addition, you can test your installation and play around with a few included examples.

Testing

To make sure everything is working correctly and Sapsan was installed without issues, run:

sapsan test

Running Examples

To get started and familiarize yourself with the Jupyter Notebook interface, feel free to run the included examples (CNN, PICAE, PIMLTurb on 3D data, PIMLTurb1D on 1D data, and KRR on 2D data). There is also a notebook with examples of plotting routines and ML network visualization. To copy the examples, type:

sapsan get_examples

This will create a folder ./sapsan_examples with appropriate example jupyter notebooks.

Custom Projects

In order to get started on your own project, proceed as follows:

sapsan create --name {name}

where {name} should be replaced with your custom project name. This will result in creation of the following file structure:

Project Folder:             {name}/
Data Folder:                {name}/data/
Estimator Template:         {name}/{name}_estimator.py
Jupyter Notebook Template:  {name}/{name}.ipynb
Docker Template:            {name}/Dockerfile  
Docker Makefile:            {name}/Makefile  

This structure allows you to focus on the designing your network structure itself in {name}_estimator.py. At the same time, you can quickly jump into Jupyter Notebook and start running your custom setup. Lastly, Dockerfile is already pre-filled to easily share your work with your collaborators or as part of a publication.

Graphical User Interface (GUI) - beta

In the aim to provide a user-friendly experience, best suited for demonstrations of your models at talks and conferences, while attempting to not sacrifice too much on customization we have designed a GUI for Sapsan. By utilizing Streamlit, a python library to build web applications, Sapsan can be fully interacted in the browser running locally. A user can tweak the parameters, edit the portion of the code responsible for the ML model, perform visual layer-by-layer analysis, train/validate, analyze the results, and more.

Lastly, Sapsan can be tried out in the demo-mode directly on the website - sapsan.app. There, one has limited editing capabilities but can explore the hyper-parameters and get a general understanding of what the framework is capable of.

Offline

sapsan.app is temporarily offline while transitioning to a new hosting service. Please refer to local GUI example for the demo.

Running GUI

In order to run it type in the following and follow the instructions - the interface will be opened in your browser

sapsan get_examples
streamlit run ./sapsan-examples/GUI/st_intro.py

Learn more at GUI Examples.

Troubleshooting

If you encounter the following error when launching streamlit

upper limit on inotify watches reached!

Then follow the following instruction by Shivani Bhardwaj to increase the watchdog limit (it won't hog your RAM)