Synapse provides the ability to record [provenance ](https://docs.synapse.org/articles/provenance.html)of experimental results. The main idea is just to keep track of inputs and outputs across multiple experiments we may do while doing analyses. Synapse has a built-in framework for this task. For example, for deep learning training during the Hackathon, a best practice will be to record the packages used, configuration parameters, inputs and model results for each training run, with artifacts recorded in Synapse using the provenance functions. I became sensitized to this issue when watching [an advertisement](https://www.youtube.com/watch?v=u90TbxK7VEA&feature=youtu.be) for [Weights and Biases](https://app.wandb.ai/), a web-based tool for recording results of deep learning training runs. Their tag line is "Those who don't track training are doomed to repeat it." The main thing they seem to be doing is recording the data that would be stored in a [Tensorboard TF Events file](https://www.tensorflow.org/guide/summaries_and_tensorboard), and providing a web viewer for that. We can also save our events files and just used Tensorboard itself to display them for review.

Created by Lars Ericson lars.ericson
Hi Lars, My sense is that the discussion forum and Slack serve different purposes. The discussion forum will work better for longer running conversations with discrete topics that may be worth saving for posterity ( If you want a discussion that can be linked to and viewed later, ie link to a discussion in a wrap-up summary), the discussion board is probably a better way to go, whereas Slack will be better for inter-intra team quick conversations, troubleshooting, and having one-on-one messaging with some mentors that will be remote as well as for teams that have local and remote members. But in the end it will depend on what most people gravitate towards. Cheers, Robert
Note I just got email for Slack so with two communication channels, it's not clear whether people will use this Discussion board much or do most of their talking on Slack. Also with the compressed timeline it's not clear people will acculturate to Synapse provenance and docker registry. Are there longer-running projects that use Synapse, and how do they come about and get organized? If the idea is open science, how much open science is actually getting done, versus the usual competitive and proprietary science? For example the professor who was my PhD thesis adviser back in the day is big on patents and told me he has to talk to a University lawyer before giving an outside talk, for fear of giving away some imminently patentable train of thought.
I highly recommend this! Provenance makes it nice and easy to keep track of what code goes where. On a similar note, it's also worth considering packaging one's analyses in a Docker container, as we have done with the walkthrough notebooks. This is a really nice way to make sure that others can reproduce your work (same libraries, OS, languages etc), and creates a permanent environment that others can use build upon your work. If you store the containers on a platform like [dockerhub](www.dockerhub.com) or the [Synapse Docker Registry](https://docs.synapse.org/articles/docker.html), it becomes very easy for others to quickly pull and work with your container. I and the other organizers would be happy to help people do this at the event.

Experiment tracking with Synapse provenance features page is loading…