Forecasting Using Python with Deep Learning

[jcbsorensen]

Hello Everyone,

I have been testing AAI SSE with python. While it seems like an extremely flexible tool, I am not sure it can provide what I seek. I was hoping some of you could answer that question.

Currently, I have been able to:

1. Use LOAD ... EXTENSION ... in the data load editor with success
2. Basic custom made functions used in expressions editor

I am mainly interested in using this extension to make forecasting using more advanced models than common ARIMA. What I am most likely going to be using are deep learning models such as long short-term memory models (LSTM). These type of models can provide predictions/forecasting with several fields as an output. Thus, the output might be 3 fields (e.g. date, shop, sales count) and 30 rows (e.g. 1 row for each future day). 

An example of Actual data:

Date	Shop	Sales
16-12-2019	Shop A	55
16-12-2019	Shop B	54
17-12-2019	Shop A	87
17-12-2019	Shop B	65
18-12-2019	Shop A	23
18-12-2019	Shop B	53
19-12-2019	Shop A	12
19-12-2019	Shop B	89

An example of Predicted data from model:

Date	Shop	Sales
20-12-2019	Shop A	55
20-12-2019	Shop B	54
21-12-2019	Shop A	87
21-12-2019	Shop B	54
22-12-2019	Shop A	38
22-12-2019	Shop B	25

All the forecasting examples I have seen so far, are only using ARIMA which provide forecasting with 1 field as an output (e.g. sales). The several fields as an output from the deep learning models seem to be an issue when the output is provided by using a custom function in the expression editor in qlik - fx as a line in a linechart. 

The single output is also seen in the video tutorial published by Qlik:

So to wrap this up with my questions:

• Can a custom python function that is called in a Qlik expression return several fields at once to Qlik?
  • If so can all the fields be displayed in a table (e.g. like the prediction table above)? or will it only work for specific charts?
  • As an example by using the data in the tables above; If I am creating a line chart using sales as a measure and shop as a dimension into several lines. Will I be able to use the predicted data to create forecasting in this linechart?

Thank you so much for your time. 

[ZYJ]

in the example of EchoTable_3 functions, it seems that several results have been echoed back, Haven't tried it by myself. 

A return row can have mutiple Duals DataType

https://github.com/qlik-oss/server-side-extension/blob/master/docs/SSE_Protocol.md#qlik.sse.Row