Add a method to add the quantiles in a timseseries graph

[lordofthejars]

As we talked about in DevNexus, I'd love to plot some time series where I set the mean and the quantiles. What I've got is an array of floats for each part, here I pasted you a real example.

==== target =====
[112.0, 118.0, 132.0, 129.0, 121.0, 135.0, 148.0, 148.0, 136.0, 119.0, 104.0, 118.0, 115.0, 126.0, 141.0, 135.0, 125.0, 149.0, 170.0, 170.0, 158.0, 133.0, 114.0, 140.0, 145.0, 150.0, 178.0, 163.0, 172.0, 178.0, 199.0, 199.0, 184.0, 162.0, 146.0, 166.0, 171.0, 180.0, 193.0, 181.0, 183.0, 218.0, 230.0, 242.0, 209.0, 191.0, 172.0, 194.0, 196.0, 196.0, 236.0, 235.0, 229.0, 243.0, 264.0, 272.0, 237.0, 211.0, 180.0, 201.0, 204.0, 188.0, 235.0, 227.0, 234.0, 264.0, 302.0, 293.0, 259.0, 229.0, 203.0, 229.0, 242.0, 233.0, 267.0, 269.0, 270.0, 315.0, 364.0, 347.0, 312.0, 274.0, 237.0, 278.0, 284.0, 277.0, 317.0, 313.0, 318.0, 374.0, 413.0, 405.0, 355.0, 306.0, 271.0, 306.0, 315.0, 301.0, 356.0, 348.0, 355.0, 422.0, 465.0, 467.0, 404.0, 347.0, 305.0, 336.0, 340.0, 318.0, 362.0, 348.0, 363.0, 435.0, 491.0, 505.0, 404.0, 359.0, 310.0, 337.0, 360.0, 342.0, 406.0, 396.0, 420.0, 472.0, 548.0, 559.0, 463.0, 407.0, 362.0, 405.0, 417.0, 391.0, 419.0, 461.0, 472.0, 535.0, 622.0, 606.0, 508.0, 461.0, 390.0, 432.0]
==================

==== prediction =====
[442.40707, 442.26746, 473.85284, 503.2582, 544.3242, 631.8106, 703.4895, 682.01056, 584.2869, 496.65305, 458.26706, 486.83524]
================================

==== Quantile 50% =====
[441.90524, 438.94196, 475.58868, 504.77777, 546.2015, 631.03503, 703.1645, 684.6702, 585.4043, 497.68643, 456.81332, 487.01727]
================================

==== Quantile 90% =====
[470.60236, 472.7702, 506.5765, 537.26117, 573.006, 667.6558, 739.82074, 709.30133, 611.8065, 520.3827, 486.74255, 515.27356]
================================

The target is the real values (so it is the past) and should be plotted from the initial of the graph until one point after that point, when the prediction starts. The prediction is the array of the mean prediction, which is the main line, and then two quantiles.

Python does this, and the output graph looks like this:

https://d2kv9n23y3w0pn.cloudfront.net/static/README/forecasts.png

It's not necessary to be the same. Also notice that in that graph there is 3 predictions, in my case with only one prediction is enough

:)

[HanSolo]

Short question, where is the time in the data?

[lordofthejars]

Oh, it is a YearMonth object, of the first series, and then there is a String that represents the frequency, so for example YearMonth 01-1964 and freq M means that from value to value there is one month. Of course the freq can be seconds, minutes, hours, days, months, ...

So if you want to do it generic enough, I would say not using YearMonth object but using the Temporal object and calling plus method setting the TemporalUnit depending on the string freq

[HanSolo]

Hmm...ok...was playing around with it and here is a first test. Question is if I do it right. The area defined by Quantile 50 and Quantile 90 is at the moment calculated by going from the prediction y value +/- quantile50 y value and the same for the Quantile 90 data. This might be wrong...

[lordofthejars]

I think it is right, I compare this graph with the one I passed which is using the same dataset, and the green part is similar to this one, when I call the quantile function it returns what I showed you, not relative to the mean but the absolute number.

Look the method I call for returning the array I pasted before is coming from forecast.quantile(0.90f)

[HanSolo]

There is one more question which is, how do I figure out the time of the prediction? It is just a set of numbers. At the moment I just moved the data manually to overlay the prediction.

[lordofthejars]

It is the same elapsed of time as the original set. So if from target data1 to data2 is 1 month, then in the prediction from pred1 to pred2 is also 1 month. So in this case if there are 12 numbers in the prediction is 12 months

[HanSolo]

Could you provide the data for all 3 predictions with their start timestamp and the time step between the datapoints. And also the start timestamp for the target data? I would like to recreate the complete chart. 😁

[lordofthejars]

Initial date for target: 1949-01

Target values: 112.0, 118.0, 132.0, 129.0, 121.0, 135.0, 148.0, 148.0, 136.0, 119.0, 104.0, 118.0, 115.0, 126.0, 141.0, 135.0, 125.0, 149.0, 170.0, 170.0, 158.0, 133.0, 114.0, 140.0, 145.0, 150.0, 178.0, 163.0, 172.0, 178.0, 199.0, 199.0, 184.0, 162.0, 146.0, 166.0, 171.0, 180.0, 193.0, 181.0, 183.0, 218.0, 230.0, 242.0, 209.0, 191.0, 172.0, 194.0, 196.0, 196.0, 236.0, 235.0, 229.0, 243.0, 264.0, 272.0, 237.0, 211.0, 180.0, 201.0, 204.0, 188.0, 235.0, 227.0, 234.0, 264.0, 302.0, 293.0, 259.0, 229.0, 203.0, 229.0, 242.0, 233.0, 267.0, 269.0, 270.0, 315.0, 364.0, 347.0, 312.0, 274.0, 237.0, 278.0, 284.0, 277.0, 317.0, 313.0, 318.0, 374.0, 413.0, 405.0, 355.0, 306.0, 271.0, 306.0, 315.0, 301.0, 356.0, 348.0, 355.0, 422.0, 465.0, 467.0, 404.0, 347.0, 305.0, 336.0, 340.0, 318.0, 362.0, 348.0, 363.0, 435.0, 491.0, 505.0, 404.0, 359.0, 310.0, 337.0, 360.0, 342.0, 406.0, 396.0, 420.0, 472.0, 548.0, 559.0, 463.0, 407.0, 362.0, 405.0, 417.0, 391.0, 419.0, 461.0, 472.0, 535.0, 622.0, 606.0, 508.0, 461.0, 390.0, 432.0

Initial date of prediction: 1961-01

Quantiles:

==== Quantile 50% =====
[441.90524, 438.94196, 475.58868, 504.77777, 546.2015, 631.03503, 703.1645, 684.6702, 585.4043, 497.68643, 456.81332, 487.01727]

==== Quantile 90% =====
[470.60236, 472.7702, 506.5765, 537.26117, 573.006, 667.6558, 739.82074, 709.30133, 611.8065, 520.3827, 486.74255, 515.27356]

Freq: M

[HanSolo]

I will implement the prediction as overlays to a line chart, meaning to say you will be able to add multiple predictions to an existing line chart.

[lordofthejars]

Oh that's great Missatge de Gerrit Grunwald ***@***.***> del dia dv., 12 d’abr. 2024 a les 13:37:

…

I will implement the prediction as overlays to a line chart, meaning to say you will be able to add multiple predictions to an existing line chart. — Reply to this email directly, view it on GitHub <#124 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALSMYK6GWHPYWGSA5YKVQ3Y47BOJAVCNFSM6AAAAABGCB2Q66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJRGU4TCMBWHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[lordofthejars]

oh btw I will write an article for Red Hat Developers Blog about your library, so maybe you can expect an increase of traffic :P Missatge de Alex Soto ***@***.***> del dia dv., 12 d’abr. 2024 a les 13:41:

…

Oh that's great Missatge de Gerrit Grunwald ***@***.***> del dia dv., 12 d’abr. 2024 a les 13:37: > I will implement the prediction as overlays to a line chart, meaning to > say you will be able to add multiple predictions to an existing line chart. > > — > Reply to this email directly, view it on GitHub > <#124 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AALSMYK6GWHPYWGSA5YKVQ3Y47BOJAVCNFSM6AAAAABGCB2Q66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJRGU4TCMBWHE> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> >

[HanSolo]

Made some progress, could you also provide the data for the other 2 predictions (the green and blue one) in the original chart? This is how it looks like at the moment.

You will find the demo in the jdk21 branch in the tests in TimeSeriesPredictionTest.java and you can run it by calling the TimeSeriesPredictionTestLauncher.java

[lordofthejars]

Oh looks great. I don't have them as my prediction was for only 12 months. And the model I am using is trained only for 12 months. If you want to try with more, I'd suggest copying and pasting the previous ones. I know it does not make much sense from the point of view of AI, but for display, I think it is ok.