Forecasting as a service - Add intelligence to your products

get it



You need to become a Contributor to join the discussion - Find out how.
João NogueiraMaker@jnunonogueira · Product @ Whitesmith
Hi Product Hunters, Have you ever seen "Back to the Future" and dreamed about knowing how the future will be? We are far from knowing who will win the next World Series or the World Cup, but we can help you forecast some very cool things! We built Unplugg for people who want an automated Forecasting API for time series data. You can use it both for your hacking project (to forecast the temperature in your room) or your awesome business (to forecast sales or energy consumption of your business). Unplugg is automated, easy to use through our API and we never keep any data. In Unplugg's website you can test directly our API by using the default data we give, introducing your own data, or you can just request an API key to integrate Unplugg in your project and explore its capabilities. The good news, you can use our testing API unlimited in your projects! Yep, you just have to ask for your API Key using our website. Thanks for checking out Unplugg, I'm really thrilled to hear your thoughts guys! Feedback and ideas for our future versions are very welcome, and we surely are available to help you using Unplugg either for personal or commercial uses! Cheers
Kelly Kuhn-Wallace@kkdub · 🎯Strategist for Hire
Can you tell us a little more about the algorithm that is used in Unplugg? How specific is it and what would a minimum v. ideal record count be? (Time series data is the most common data set most firms have, but I see a lot of challenges with sample size and collection size coming into it.)
Miguel TavaresMaker@mgontav · Data Guy, Whitesmith
@kkdub We can't go deep into details on how the algorithm works, but at it's core we are using SARIMA models (so no ML/deep learning approach, but statistical methods) combined with a lot of previous analysis and pre-processing of the input data to find out sane parameters for the forecast. Along with this there is some manual optimization done previously by us on various datasets, to figure out optimal parameters for all the processing steps. As far as specificity goes, we can't make claims that it is completely generic - that would be quite a feat. We started by tailoring it to a specific use-case - energy consumption forecast - but soon figured out that the resulting models were not really reflecting energy consumption patterns, but human-driven patterns, that tend to follow similar trends and seasonalities whatever the domain is. So right now we're opening it up for testing on similar human-centric domains - usage of appliances/tools, website visits, retail metrics... obviously it won't work equally well on every domain, but that's what we are here to figure out. Sample size: that depends heavily on the nature of the data, granularity and seasonality - we tended to find good results while testing by supplying it with at least 3 seasonal repetitions to forecast the next one (eg. three weeks of data to forecast the next one) but of course that would need to change if, for example, we want to take into account monthly/yearly fluctuations. It ends up being a matter of testing it with the available data and evaluating the results.
Kelly Kuhn-Wallace@kkdub · 🎯Strategist for Hire
@mgontav Super helpful -- ty!!
Bryan Dickens@brdick · PM, Microsoft
forecasts can also come with confidence levels, is there any way to get a std dev or 25%/75% quartile ranges on these predicted values?
Miguel TavaresMaker@mgontav · Data Guy, Whitesmith
@brdick Not yet, unfortunately. We're working on making that available ASAP as it is a common request in the feedback we're having. In our next release (1-2 weeks) we're hoping to make it available.
Daniel F LopesHunter@danflopes · I help build products faster @Whitesmith
Hi everyone, Say hi to Unplugg. It's free to use, and the Unplugg team is anxious to hear about your feedback and cool experiences with it.