Possibility of a heroku ESS implementation?

[SkotBotCambo]

Hello All, Thank you for making this great project open-source. There are a great many similar projects (AWARE, WISDM, Passive Data Kit), but this one seems to be the most suitable and well thought out for the academic community. I am considering forking this repo and creating a version of ESS that can be easily spun up as a Heroku instance. Please let me know if you would like to coordinate or if you have any advice or see any major challenges associated with such a project.

Cheers! Scott Cambo

[yvaizman]

Hey Scott.

I am glad you discovered this work and this code and that you are interested in working with it. I will not be available to heavily maintain the code, so you are free to fork the repo and publish your revised code.

I am not familiar with working with Heroku, so I don't know about any expected challenges :-) The ESS should provide a web interface with 2 basic HTTP API functions (one to upload sensor data zip file for a timestamp and another to report context-labels for a timestamp), so you just need that interface to remain. I'm guessing it should be simple.

For the backend, it involves saving data files (unless you are interested in real-time context-predictions only) - perhaps you can make the saved files more elegantly collected by the researcher (of course, while maintaining strict security and allowing only authorized people access the data).

Good luck. Please update! Thanks, Yonatan.

[SkotBotCambo]

Awesome, thanks for the quick response, Yonatan. We are going to look at it over our (Northwestern University) spring break to see what we are getting ourselves into, but as you said it should be fairly straight forward. Thanks again and great work! :)

[yvaizman]

also, I think it is convenient to have a core of python code on the server - it made it easier for me to use some code pieces that I already used for my offline experiments: for feature extraction and for implementing the neural network to produce the probability predictions.

The feature extraction python code that I used also assured that the extracted features were consistent with the publicly available sensor-features from the ExtraSensory Dataset (http://extrasensory.ucsd.edu/#download).

So maybe you'd like to (I'm not sure if it's possible) keep the python core and implement the surroundings.

[SkotBotCambo]

That sounds like a good idea. Is this the same code that is in the tutorial IPython Notebook?

[yvaizman]

I don't think so. The tutorial on http://extrasensory.ucsd.edu/#tutorial processes the published data files (the already extracted features), so it doesn't include the feature-extraction code. As for the classification, I think in the tutorial there is a simpler model (SVM maybe?) and not the neural networks that I used in my paper (and whose implemented in the ESS code). Yonatan.

On Wed, Mar 21, 2018 at 1:26 PM, Scott Allen Cambo <notifications@github.com

wrote:

That sounds like a good idea. Is this the same code that is in the tutorial IPython Notebook?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/cal-ucsd/ExtraSensoryAndroid/issues/1#issuecomment-375084175, or mute the thread https://github.com/notifications/unsubscribe-auth/AB8c1N5jHyNwaNzkDPE0GJD_vgyzc7AGks5tgreQgaJpZM4S1l8n .

[SkotBotCambo]

Ah right, that makes sense. These must be the files that are in ESA_components/ESS/Classifiers?

[yvaizman]

Correct :-)

On Wed, Mar 21, 2018 at 1:36 PM, Scott Allen Cambo <notifications@github.com

wrote:

Ah right, that makes sense. These must be the files that are in ESA_components/ESS/Classifiers?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/cal-ucsd/ExtraSensoryAndroid/issues/1#issuecomment-375086895, or mute the thread https://github.com/notifications/unsubscribe-auth/AB8c1KbWG8-HIs0-1jVdrBkHY1_CM0lKks5tgrnCgaJpZM4S1l8n .