Related Posts
Chief
Any idea of ericcson project in TCS
Hi,
I'm currently having an offer from BlueOptima and Cohesity and am conflicted between the two.
I have offers in the SDET profile.
Cohesity Inc is providing me MTS and @BlueOptima is providing SDET-1.
Glassdoor reviews more of less place the two companies almost equally, and the package being offered by the two is almost in the same range.
My preferences include:
Learning opportunities in the role.
Company's work environment.
Company's growth prospects, are also a consideration.
YOE: 2 years
What’s the culture at NPR?
Rising Star
Additional Posts in Data & Analytics Consultants
Bowl Leader
New to Fishbowl?
Download the Fishbowl app to
unlock all discussions on Fishbowl.
unlock all discussions on Fishbowl.
This is a classic time series analysis problem. And it's not really multiple "classes" - just separate models. You would get data for each store, and fit a time series model to each store's data. https://otexts.com/fpp2/
I'll expand, and take your example as just a toy example. Let's say you're forecasting for 100 stores across the country. In this case, you'll want to understand the underlying business problem: are you trying to provide some kind of aggregate forecast for execs? Or a granular forecast for store managers?
If the former, generating store-level forecasts and aggregating them is going to capture too much of the noise within each store's data, and possibly compound the error within each store's model. The end result is a high-variance model that feels useless.
If the latter, you may still want to look at doing some aggregation (e.g. region-level forecasting) in case you have missing data, lack of data, extreme events (e.g. storms), etc.
2 ways. 1st build 12 models each model for 1 store. 2nd build 1 model with a 12 elements vector depend variable
True, a multivariate time series would work, though IIRC it only makes sense if the variables (the different stores' sales) have some relationship to each other. That could be true if they're located near each other since one store's sales might cannibalize another's.
The good use-case might be try to create dummy variables by using Store ID (if there’s any). It can help you to save loads of time - no need to create separate models for each store. Then think about what type of independent variables that you have in your hand, checking correlation levels. Consider to use dynamic regression model with time series or simply create day of year variables and fit into the regression model. AICs & RMSE can be used as the performance trackers