Gsoc project 2021

I’m an IT & Maths major at Cluster Innovation Centre, Delhi University. I’m interested in contributing to the Time series models project. I’m familiar with general computing libraries in python(pandas, numpy, scikit learn, scipy etc), jupyter notebook , git, matplotlib etc. I’m also familiar with general regression models, I’ve worked on a handful ML projects - including a mathematical modelling (COMAP 2020) of temperature over time among pelagic fishes my project was the only one to get an honourable mention from India. I’d love to work on this, it would be great if you could help me get started, maybe suggest me things to work on? I’ve talked about this to one of the mentors( Ravin Kumar) and he suggested me to bring this up here.

Thank you
Meenal Jhajharia
Sophomore Year - BTech(IT & Maths)
Cluster Innovation Centre
University of Delhi

2 Likes

Welcome Meenal!
Excited to chat with you here as well. Thanks for the correspondence over email and excited for your interest in GSOC, even if for projects other than PyMC and ArviZ!

1 Like

Hi

I just began writing the Proposal and I made a very brief draft for the abstract, I don’t want to write a detailed proposal in the wrong direction. I’d be grateful if any of you could have a look and let me know if this works, if yes then I’ll go ahead and write the whole thing and then ask for reviews. Else, I’ll make changes accordingly. I hope I’m not causing much inconvenience. I have attached the google doc link, where you can comment, or reply here. Thank you for your time.

Regards
Meenal

Google Doc - Proposal Abstract

I’m really looking forward to this, I’ve been reading a few papers on ARIMA and time series models in general and I’m excited to bring in some sort of amalgation of Probability theory and computing that somewhat benefits everyone? As in, for researchers working on something that needs a forecasting model, if they can use PyMC3 to not just compute, but also understand what’s happening better then win-win!! Right now, all my ideas are really abstract and I don’t wanna come off as too ambitious and unrealistic, I actually trimmed down 2-3 points in the abstract and I’ll try to pin point my contributions and plan further in the proposal. I might sound like I’m trying to do everything at once, but I feel like the arviz-pymc3 time series projects can actually go hand in hand, especially for the two newer models, but that’s for later, if a possibility. I don’t mind working on that outside of GSoC as well!!

1 Like

Thanks for sharing your interest Meenal! We’re looking forward to seeing how your proposal develops!

1 Like

Hi Meenal,
In regards to the time series project there for pymc3 an internal discussion about whether it would end up being a good GSOC project or not. To be safe hold off on working on it for 24 hours or so we don’t accidentally end up wasting your time.

In the meanwhile though the ArviZ time series project is still possible!

1 Like

Okay, thankyou for letting me know, I hope things work out in favour as I’m really excited to do this.

@almostmeenal if not let us know if youre interested in the ArviZ Time Series project. It will be very close in subject matter to this one, more on the diagnostic and visualization side! It also means we can reuse parts of the proposal you already started writing quite easily

2 Likes

Hi Ravin, yes that sounds great!! I’ll try to center my proposal research around generic time series stuff for now, and as soon as a decision is made I’d be happy to work on either of them. Ideally the pymc3 one, but the Arviz project sounds equally great as well.

Here’s the update: PyMC3 could use a time series project. We’ve updated the description to remove prophet and add support for state space models. Sorry for the delay!

1 Like

glad to hear that, thanks a lot for letting me know!! I’ll get to the proposal immediately

Hi everyone, I’m halfway through the proposal, this is a very rough draft of the first 50-60 percent (another 3-4 pages need to be added, i.e. Timeline, Possible problems etc). I’d be really grateful if any of you would have a look and suggest improvements, or point out mistakes or something that I totally missed out, or if I’m entirely writing in the wrong direction or something.

1 Like

I’m done with the first draft of the entire proposal, I would be so grateful to get reviews and suggestions now. The link for the proposal is the same as above!!!

tagging relevant project mentors @RavinKumar @twiecki

1 Like

Following on the Bayesian Estimation Supersedes the T-Test — PyMC3 3.11.2 documentation example, there we build a model to estimate both group means, and then we evaluate if the mean of one group is larger than the other. In general, we want to focus on estimation rather than testing, i.e. we don’t run a test to see if the means are equal or not and then build a model to estimate the mean (or means) depending on the result of the test. Here I think a similar situation should apply, either we set these parameters to a fixed value or we estimate them as part of the model somehow. From bayesforecast/Sarima.R at master · asael697/bayesforecast · GitHub it looks like setting the parameters as fixed is common practice. Again note that timeseries is not my area of expertise and I have never used nor an arma nor an arima model, take all of this with a grain of salt.

1 Like