Re:Yhat Sciencebox
by rguldener on Thursday June 19, 2014 @09:52AM
(#7916044)
I think this is cool but for me personally it would be so much cooler if it was open-source. Sure I understand they want to make money with it but really this doesn't offer me enough more than my current setup that I would be willing to pay $0.05 - $1.40 per hour. That's up to $33 per day or $1041 per month, (quite) pricey for some background execution and rsync functionality.
My current setup is to use a workhorse server where I set up a samba share for my home folder (so I can edit scripts remotely from my notebook without the git-push-pull dance) and a remote iPython notebook. Both config files are write once and forget and I get 90% of their functionality built in (without the web GUI). Would love to learn about other people's setup though, I am sure mine can be improved.
Re:Yhat Sciencebox
by fyrabanks on Thursday June 19, 2014 @11:24AM
(#7916578)
Assuming the upper end of that--for a lab with 10 researchers (forget the students, postdocs, etc. for a second), you could save maybe half the cost by hiring a single IT person to do this. Aside from their sb client, I don't see anything in this that can't be downloaded and installed (either from source or automagically from any of the freely available science-related R, Python, etc. repos).
Neat idea, I just find it difficult to justify a $1000/mo per-machine cost, especially in academia.
Re:Yhat Sciencebox
by rguldener on Thursday June 19, 2014 @10:12AM
(#7916143)
True, but if you only need a small instance you can also just use your laptop to do the processing and do away with a remote machine altogether. You are right though $30 a month for a good management tool seems much more reasonable, too bad they tie the price to the server performance.
Re:Yhat Sciencebox
by larrydag on Thursday June 19, 2014 @11:50AM
(#7916720)
you would think a Github like model would work. If the project is open source then its free but if you want to deploy an enterprise-wide or private product then you would pay the server costs.
Re:Yhat Sciencebox
by elliott34 on Thursday June 19, 2014 @10:43AM
(#7916311)
This pairs well with the recent AirBNB fraud detection article. Airbnb has a lot of people to construct a robust machine learning architecture. For small start ups trying to make a recommendation engine, why not do sciencebox+ yhat cloud/enterprise? Especially for data science people that are don't have great dev skills.
Re:Yhat Sciencebox
by evancasey on Thursday June 19, 2014 @07:32PM
(#7918852)
Seems like a useful tool that would make doing data analysis on a remote machine much easier than with plain old AWS + anaconda.
However, the incremental improvement of renting an EC2 instance (even with 8 cores and 65gb) pales in comparison to using a distributed data processing approach with medium to large datasets. Any plans for supporting Hadoop/Spark + Amazon EMR in the future?
Re:Yhat Sciencebox
by radikalus on Thursday June 19, 2014 @11:51AM
(#7916722)
Two things:
- GPU instance support would be really really helpful
- Pricing of the higher perf tiers is a bit painful which changes the normal AWS $ optimization metrics pretty massively
Otherwise -- it's definitely something we'd consider using; I end up doing 99% of my work on my laptop because AWS setup fixed cost time is 30minutes plus every time.
Re:Yhat Sciencebox
by cschmidt on Thursday June 19, 2014 @10:19AM
(#7916173)
Statwing is trying to do nice visualizations and analysis in a web browser, without programming. This is more of a way to use R or python to do data analysis on a remote AWS machine. You still have to know what you're doing in R or python, but it makes the "remote" part easy.
Re:Yhat Sciencebox
by ehurrell on Friday June 20, 2014 @03:33AM
(#7919849)
I had thought Statwing set up the remote stuff and then automated a lot of tasks in-browser, but it doesn't seem to give the easy access this does. This is a lot more attractive!
My current setup is to use a workhorse server where I set up a samba share for my home folder (so I can edit scripts remotely from my notebook without the git-push-pull dance) and a remote iPython notebook. Both config files are write once and forget and I get 90% of their functionality built in (without the web GUI). Would love to learn about other people's setup though, I am sure mine can be improved.
Neat idea, I just find it difficult to justify a $1000/mo per-machine cost, especially in academia.
However, the incremental improvement of renting an EC2 instance (even with 8 cores and 65gb) pales in comparison to using a distributed data processing approach with medium to large datasets. Any plans for supporting Hadoop/Spark + Amazon EMR in the future?
Otherwise -- it's definitely something we'd consider using; I end up doing 99% of my work on my laptop because AWS setup fixed cost time is 30minutes plus every time.
And perhaps updating the COBOL scripts from 1970-something that I inherited