Machine Learning

This document describes how to prepare & execute the machine learning demo

Prerequisites
Demo Preparation
- Pre Demo Checks
- Deploy OpenDataHub with a JupyterHub in the Manuela-ML-Workspace
Demo Execution
- Demo ML modeling

Prerequisites

The demo based environment have been bootstrapped.

See Machine Learning based Anomaly Detection and Alerting

Demo Preparation

The demo prepation is focusing on the Data Scientist workbench which is a OpenShift powered Open Data Hub.

Environments:

OCP3: Predeployed Manuela-ML-Workspace
OCP4: Deploy Manuela-ML-Workspace during the demo (optional)
CRC: Deploy the Manuela-ML-Workspace in you own environments for any use case

This guide assumes that you are using the stormshift environment.

Pre Demo Checks

Ensurure that the Manuela-ML-Workspace is deployed and functioning on OCP3
Delete the Manuela-ML-Workspace of OCP4 in case you like to show the deployment.

Deploy OpenDataHub with a JupyterHub in the Manuela-ML-Workspace

The Manuela-ML-Workspace should exists or be deployed on OCP3 before running the demo. You can deploy the Manuela-ML-Workspace during the demo on OCP4 or your CRC in case you like to show how to deploy the Open Data Hub and upload the Jupyter and training data.

Let’s use ocp4 during the following steps. Login into ocp4 as admin or with admin privileges:

oc login -u XXX -p XXXX --server=https://api.ocp4.stormshift.coe.muc.redhat.com:6443

Please clone the manuela repository into your home directory. This repo contains everything required to set up the demo. You can choose a different directory, but the subsequent docs assume it to reside in ~/manuela-dev .

cd ~
git clone https://github.com/sa-mw-dach/manuela.git

Deploy the OpenDataHub Operator

Starting with ODH V0.6, it is a global operator. To deploy it, use

cd  ~/manuela/namespaces_and_operator_subscriptions/opendatahub
oc apply -k .

If oc apply -k . fails, please try kustomize build . | oc apply -f -

Wait a minute until the operator is running:

oc get pods -n openshift-operators
NAME                                    READY   STATUS    RESTARTS   AGE
opendatahub-operator-546d49d59b-qd8hz   1/1     Running   0          48s

Deploy a OpenDataHub instance with a JupyterHub in the Manuela-ML-Workspace namespace

cd  ~/manuela/infrastructure/opendatahub
oc apply -k . 

Follow the instantiation of the jupyterhub in the OpenShift Console or using oc:

oc get pods -w -n manuela-ml-workspace

Launch a Jupyterhub

Get the Jupyterhub Url either from the route in the OpenShift Console or using oc::

echo https://$(oc get route jupyterhub -o jsonpath='{.spec.host}' -n manuela-ml-workspace )

https://jupyterhub-manuela-ml-workspace.apps.ocp4.stormshift.coe.muc.redhat.com

Login with OpenShift credentials
Spwan a notebook s2i-minimal-notebook:3.6 using the defaults
Upload Anomaly-Detection-simple-ML-Training.ipynb and raw-data.csv from ~/manuela-dev/ml-models/anomaly-detection/

launch-jupyter

Instead of uploading the files via the Jupyter UI from your local computer, you can start a Jupyter Terminal and download direclty from git:

curl -O https://raw.githubusercontent.com/sa-mw-dach/manuela-dev/master/ml-models/anomaly-detection/Anomaly-Detection-simple-ML-Training.ipynb

curl -O https://raw.githubusercontent.com/sa-mw-dach/manuela-dev/master/ml-models/anomaly-detection/raw-data.csv

Demo Execution

Demo ML modeling

Demo the notebook

Open the notebook Anomaly-Detection-simple-ML-Training.ipynb

Option 1: Lightweigt demo

All output cells are populated. Don’t run any cells.
Walk through the content and explain the high level flow.

Option 2: Full demo

Clear current outputs: Cell -> All Output -> Clear
Run each cell [Shift][Enter] and explain each step.

Demo model serving

For keeping the demo setup simple, lets use for show the model serving.

Show the running seldon pods in manuela-stormshift-messaging.

oc get pods -n  manuela-stormshift-messaging | grep 'seldon\|anomaly'

anomaly-detection-predictor-0-anomaly-detection-796887f9899c2jj   2/2     Running   0          22h
seldon-controller-manager-76d49f78b9-k7xc7                        1/1     Running   0          25h

Test the anomaly detection service

Make a test call:

curl -k -X POST -H 'Content-Type: application/json' -d '{"data": { "ndarray": [[16.1,  15.40,  15.32,  13.47,  17.70]]}}' http://$(oc get route anomaly-detection -n manuela-stormshift-messaging -o jsonpath='{.spec.host}' -n manuela-stormshift-messaging )/api/v1.0/predictions

Output:

{"data":{"names":[],"ndarray":[1]},"meta":{}}

The prediction is "ndarray":[1]. This is an anomaly.

Show logs to see anomaly-detection-predictor in action

Either on the OpenShift console or using oc

oc logs $(oc get pod -l  seldon-app=anomaly-detection-predictor -o jsonpath='{.items[0].metadata.name}' -n manuela-stormshift-messaging) -c anomaly-detection -n manuela-stormshift-messaging

Expexted result:

 Predict features:  [[ 9.58866665  8.88145877 10.60920998 11.53665955 11.65813195]]
Prediction:  [0]
2020-05-03 17:40:13,481 - werkzeug:_log:113 - INFO:  127.0.0.1 - - [03/May/2020 17:40:13] "POST /predict HTTP/1.1" 200 -
 Predict features:  [[12.59645277 13.17329123 13.91231828 15.80360728 16.36178987]]
Prediction:  [0]
2020-05-03 17:40:15,958 - werkzeug:_log:113 - INFO:  127.0.0.1 - - [03/May/2020 17:40:15] "POST /predict HTTP/1.1" 200 -
 Predict features:  [[10.66335637  9.58866665  8.88145877 10.60920998 11.53665955]]
Prediction:  [0]
2020-05-03 17:40:18,549 - werkzeug:_log:113 - INFO:  127.0.0.1 - - [03/May/2020 17:40:18] "POST /predict HTTP/1.1" 200 -
 Predict features:  [[11.21156663 12.59645277 13.17329123 13.91231828 15.80360728]]
Prediction:  [0]
2020-05-03 17:40:20,601 - werkzeug:_log:113 - INFO:  127.0.0.1 - - [03/May/2020 17:40:20] "POST /predict HTTP/1.1" 200 -