In this session you will see how to take your machine model from development to production by watching the steps involved, which include: 1) Developing a ML model crafted via a Jupyter Notebook directly on top of Kubernetes/OpenShift; 2) Publishing that model as a service to be shared with your team or even the world; and 3) Monitoring the RESTful service via Grafana.

1. Sophie Watson
sophie@redhat.com
@sophwats
Jupyter Notebooks for Machine learning
on Kubernetes and Openshift

2. Sophie Watson
sophie@redhat.com
@sophwats
Jupyter Notebooks for Machine learning
on Kubernetes and Openshift

3. Agenda
- ML workflow
- Kubernetes and OpenShift for Machine Learning
- Demo

4. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation

5. feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying
problem
and metrics
codifying
problem
and metrics

6. codifying
problem
and metrics
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying
problem
and metrics

7. feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying
problem
and metrics
codifying
problem
and metrics
data
collection
and cleaning

8. feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying problem
and metrics
codifying problem
and metrics
feature
engineering
data collection
and cleaning

9. codifying problem
and metrics
feature
engineering
data collection
and cleaning
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying problem
and metrics

10. codifying problem
and metrics
feature
engineering
data collection
and cleaning
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
data collection
and cleaning
codifying problem
and metrics

11. data collection
and cleaning
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
codifying problem
and metrics
codifying problem
and metrics
feature
engineering
model
training
and tuning
data collection
and cleaning
legit

12. data collection
and cleaning
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
codifying problem
and metrics
codifying problem
and metrics
feature
engineering
model
training
and tuning
data collection
and cleaning
legit

13. data collection
and cleaning
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation
codifying problem
and metrics
data collection
and cleaning
codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation

14. feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation
codifying problem
and metrics
data collection
and cleaning
codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
model
deployment
monitoring,
validation

15. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation

16. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation

17. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation
Robust, repeatable pipelines
Model monitoring and alerting
Scale up
Scale out
Self service environments
Reproducible environments
Multi-tenant environments
Access to specialised hardware
Scale out

18. base image
configuration and
installation recipes
user application code
979229b9
33721112
e8cae4f6
2bb6ab16
a8296f7e
a6afd91e
6b8cad3e

19. base image
configuration and
installation recipes
user application code
979229b9
33721112
e8cae4f6
2bb6ab16
a8296f7e
a6afd91e
6b8cad3e

20. base image
configuration and
installation recipes
user application code
979229b9
33721112
e8cae4f6
2bb6ab16
a8296f7e
a6afd91e
6b8cad3e

21. base image
configuration and
installation recipes
user application code
979229b9
33721112
e8cae4f6
2bb6ab16
a8296f7e
a6afd91e
6b8cad3e

22. base image
configuration and
installation recipes
user application code
979229b9
33721112
e8cae4f6
2bb6ab16
a8296f7e
a6afd91e
6b8cad3e
model in production
on 16 July 2019

28. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation
Robust, repeatable pipelines
Model monitoring and alerting
Scale up
Scale out
Self service environments
Reproducible environments
Multi-tenant environments
Access to specialised hardware
Scale out

29. In [ ]:
In [ ]:
In [ ]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
Exploring my data set
| Python 3
In [ ]: data.to_parquet("data2.parquet")

30. In [*]:
In [ ]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [ ]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [ ]: data.to_parquet("data2.parquet")

31. In [1]:
In [ ]:
Out [1]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [ ]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [ ]: data.to_parquet("data2.parquet")

32. In [1]:
In [*]:
Out [1]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [ ]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [ ]: data.to_parquet("data2.parquet")

33. In [1]:
In [2]:
Out [1]:
Out [2]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [ ]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [ ]: data.to_parquet("data2.parquet")

34. In [1]:
In [2]:
Out [1]:
Out [2]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [3]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [ ]: data.to_parquet("data2.parquet")

35. In [1]:
In [2]:
Out [1]:
Out [2]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [3]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [*]: data.to_parquet("data2.parquet")

36. In [1]:
In [2]:
Out [1]:
Out [2]:
import pandas as pd
data = pd.read_parquet("data.parquet")
data.size
120000
from plot_tool import plot
plot_df = pd.DataFrame(data, columns=["x", "y"])
plot(plot_df)
Exploring my data set
| Python 3
In [3]: new_data = pd.read_parquet("new_data.parquet")
data = data.append(pd.DataFrame(new_data))
So far we have loaded in the data and plotted it. Then we appended our new
data. Now we save it to file.
In [4]: data.to_parquet("data2.parquet")

37. Identity and RBAC
Project Isolation
Sharing of resources
Resource Isolation
Resource Quotas
Priority

38. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data collection
and cleaning
model
deployment
monitoring,
validation
Robust, repeatable pipelines
Model monitoring and alerting
Self service environments
Reproducible environments
Multi-tenant environments
Access to specialised hardware

40. @sophwats
Demo

41. codifying problem
and metrics
feature
engineering
model training
and tuning
model
validation
data
collection and
cleaning
model
deployment
monitoring,
validation
OpenShift
Pipelines

42. codifying problem
and metrics
model
validation
data
collection and
cleaning
model
deployment
monitoring,
validation
2 3
OpenShift
Pipelines

43. REST endpoint
OpenShift
Serverless