Implement some TimeSeries Analysis using the serverless engine RedisGears, Redis TimeSeries and Facebook Kats as toolkit to analyze time series data.

It shows how to:

• build a custom Redis docker image with given modules
• generate timeseries mocked data using Kats from an external python process
• define a RedisGears function to perform timeseries forecasting using Kats
• load RedisGears function from an external python process
• fetch the predictions from a (java) Spring Boot client invoking the RedisGears function

Jupiter notebook provided, containing the Kats logic used in the project: here.

docker-compose up # --build # First Time

😣 Struggled a lot trying to install KATS to be run in RedisGears venv.
The only way I was able to make it work was manually installing dependencies. :( See RedisGears Configuration

$ docker ps # Get process id
$ docker exec -ti <id> /bin/bash
# activate rg venv
$ source /var/opt/redislabs/modules/rg/.venv-/bin/activate
$ ln -s /usr/lib/x86_64-linux-gnu/libffi.so.7 /usr/lib/x86_64-linux-gnu/libffi.so.6
$ pip3 install --upgrade pip
$ pip3 install --no-cache-dir setuptools
$ pip3 install numpy pandas convertdate Cython~=0.22 pystan~=2.18
$ pip3 install fbprophet==0.7.1
$ pip3 install kats

Kats needs to be installed on python < 3.9. Create 3.8 venv and install requirements.

Run Simulator for generating time series data.

Run the python client to load RG functions.

Watch redis server log while RG functions and dependencies installation are being loaded and requested.

Spring Boot Process offers a Rest endpoint to retrieve the predicted values for the next days.

# Get Predicted values for next 3 days
curl http://localhost:8080/3
[{"t":1640995200000,"v":6068.410402495052},{"t":1641081600000,"v":5706.191453574134},{"t":1641168000000,"v":5617.7999465757375}]

The Prophet Model needs to be trained to provide accurate results