Can i create keras model as UDF on Kapacitor

In my application I would like to make a pipeline which will take timeseries data as input and returns error from keras model deployed on pipeline. Below is my requirements,,.

Input: Timeseries ( dataframe like below
timestamp,value
2017-03-05,25
2017-03-06,38
2017-03-07,23
''
''
''
Output:

{"model1":rmse,"model2",rmse,...............................}

These models are deployed on pipeline which will be running parallely.

Can somebody help me to solve this.

Start by checking Kapacitor's UDF capabilities and whether a Keras model can be invoked in a pipeline. Define how the time-series input, parallel model execution, and per-model RMSE output should be represented; the issue does not name a file, test, or existing entry point.

Written by the indexing model from the issue text.