search

biolab / orange3-timeseries

🍊 :chart_with_upwards_trend: Orange add-on for analyzing, visualizing, manipulating, and forecasting time series data.
Other
62 stars 40 forks source link

Time Series add-on - "As Time Series" widget extension of functionality #239

Closed StatViser closed 1 year ago

StatViser commented 1 year ago

Hi,

INTRODUCTION: Orange is a perfect tool to be used in education, especially to give students hands-on experience with the tools supported, keeping their focus on understanding and getting familiar with the modelling technique discussed. This also holds for the 'Time Series'-add on, that was recently updated to version 0.5.0. The add-on makes it possible to introduce time-series methods 'seamlessly' in a discussion of relevant Data Science methods!

ISSUE/REQUEST RELATED PROBLEM However, there is one point my students are struggling with when starting to analyse a time series within Orange: the definition of the 'data attribute' through the 'As timeseries' widget, as now in some cases they first have to define the data attribute within Excel before it will be easily recognized by the 'As timeseries' widget. The other alternative is to use the ''Sequence is implied by instance order'-option, but then the starting point and actual dates (eg. jan 2020 etc...) are not available for plotting.

PROPOSED SOLUTION: Therefore my request/suggestion: would it be possible/feasible to have a 3rd option for this widget, where you use the 'Sequence is implied by instance order'-option but additionally (!) a starting date and frequency of sampling is indicated (eg. start=2020.1 frequency=12 for a monthly pattern, starting at January 2020) and an additional column with an adequate 'data attribute' is generated and used for eg. the time axis in plots etc. This is the solution that is also used in R for creating a time series from a general data series, using the command "ts", eg.: ts(1:10, frequency = 4, start = c(1959, 2)) # 2nd Quarter of 1959 for a quarterly series. This would make life more easy for the students and avoid the need for Excel-use, only to create the 'data attribute' (where there might also be an issue of licensing!)

Would be nice if a solution in the direction indicated could be made available in a future version!

Thanks in advance and for all the effort given to the Orange-initiative,

Koo Rijpkema, the Netherlands

janezd commented 1 year ago

Thank you for this excellent idea! This will be really useful.

I suspect that it will be me who implements it, but I currently have several other, more urgent tasks. I promise to do it as soon as I can. It shouldn't be too difficult.

StatViser commented 1 year ago

Thanks janezd for your immediate reply: would be perfect, so looking forward to it. Especially because I want to introduce my students Data Science to Time Series ideas, as extension of their 'Data Science' toolkit! The "Time Series" add-on is helpful in this as it also covers, apart from ARIMA-models, ARIMAX and VAR models! Nice if students can be introduced to these principles in an early stage and can get 'hands-on' experiences with it.