#acumos-meeting: Architecture Committee
Meeting started by farheen at 14:12:57 UTC
(full logs).
Meeting summary
- Release Status and issues (farheen, 14:13:24)
- no blockers but there are critical and high
defects that need to be addressed. (farheen,
14:14:59)
- Epic 3256 was expecting to have the raw client
library delivered. (farheen,
14:15:58)
- Manoop worked with Priya on a model that had
issues with protobuf conversion. When they start using the new raw
client lib they will not go through the protobuf conversion. This
needs to be watched and closed. (farheen,
14:16:55)
- Guy - all the code is there. We have to sit
down with Tommy to show him how to use it. (farheen,
14:17:17)
- It's up to Guy and Tommy to get it
going. (farheen,
14:17:32)
- Murali - is there user story? (farheen,
14:17:55)
- Manoop create a task against the epic to
track. (farheen,
14:18:12)
- ACTION: Guy create a
task on Epic 3256 regarding meeting with Tommy. (farheen,
14:18:55)
- Are we able to make progress on test?
(farheen,
14:19:08)
- Ken - Test team is blocked on an issue with the
license editor to not be able to proceed. (farheen,
14:19:33)
- It is a port issue that we will have resolved
shortly. (farheen,
14:19:44)
- Access to the editor is a problem not the
edtior. It is completely separate installation. Bryan offered the
instructions on Tuesday but no one was able to set up. (farheen,
14:21:03)
- Bryan Security verification and deployment are
separate installations. Not a part of the core. It is a learning
process. (farheen,
14:21:54)
- Ken - Does each instance of Acumos need it's
own Jenkins? Can we re-use the same jenkins? (farheen,
14:22:33)
- Bryan - Each instance of acumos today has to
have it's own jenkins. In the future that will not be
required. (farheen,
14:23:43)
- Vineet - log user stories have to be
incorporated into all the components. (farheen,
14:24:43)
- Manoop - Review that they are the java objects
but using these modules will have to incorporate into the
maintenance release. We need someone can verify, re-use.
(farheen,
14:25:38)
- Vineet - Yes, we want to show what we have
done. (farheen,
14:25:58)
- ACTION: Manoop add
Vineet's demo of logs in Clio to next weeks agenda. (farheen,
14:26:29)
- Murali - Clio press release slide reminder.
Please respond to my request. (farheen,
14:26:55)
- Ken - License editor. I am finding in ML
WOrkbench as well in configuring all the ports. I understand they
want the some pluggable and non pluggable. Is there ML workbench
had to have a fully qualified name and ports. Are we going to look
at Acumos as much more pluggable. Are we going to start to look
from an architectural perspective. (farheen,
14:28:33)
- Manoop - Requires a kubernetes deployment
expert. (farheen,
14:28:48)
- Bryan components do interact intnernally.
Front door are designed to be exposed through the ingress end point.
They are presented in an iframe and their content comes from a front
end service. (farheen,
14:29:29)
- Ken - Dockerized version it doesn't matter
which Docker or K8. We have to pick a port. Service names were
said to use the fully qualified names. (farheen,
14:30:21)
- Byran - As mentioned even under docker you can
communicate outside of the cluster. Every docker is known by it's
docker name. Exact same in k8 use the cluster domain name.
(farheen,
14:31:15)
- Bryan - possible to have secure intra
connections using the service page. (farheen,
14:31:36)
- Ken - If you're talking inside the container
everyone can start using the same port. I can't test. (farheen,
14:32:08)
- Bryan - every component uses the same port at a
different service. Any component that has to have a service
exposure it has to be mapped. External mapping should be
minimal. (farheen,
14:32:46)
- Ken - What I usually get is everything needs to
be external. I don't know if someone has hard coded to the external
port. (farheen,
14:33:18)
- Bryan encourages to review the architecture
diagram. (farheen,
14:33:34)
- Ken - The architectural perspective it needs to
be reviewed by the architecture committee. (farheen,
14:35:24)
- Manoop - Follow the architecture
guidelines. (farheen,
14:35:58)
- is ML Workbench on that diagram? (farheen,
14:36:09)
- yes (farheen,
14:36:13)
- there is a core and they do need to be updated.
Some aspects have been clarified. (farheen,
14:36:33)
- Apache Nifi Integration - Nat (farheen, 14:38:25)
- IBM supported Romeo is from IBM. IBM AIF
projects to work with Acumos AI (farheen,
14:39:04)
- the models we onboard on the platform can
leverage this tool set for unbiased vs biased context. Ken worked
with design tools and the Nifi pipeline. Romeo has worked out a
prototype of Apache Nifi AI 360. Both Acumos AI and AIF will work
together. (farheen,
14:40:13)
- Romeo - sharing screen. we have a couple of
toolkits (farheen,
14:40:42)
- AI Bias tool lets you check for bias. Python
library. In order to integrate thru Nifi possibilities. Hello
fairness script but data is generated in the script and is editable
to see how the lib works (farheen,
14:42:02)
- data driven ground truth and model predictions
in the database. After that fairness call (farheen,
14:42:28)
- Able to add bias detector to the data
flow. (farheen,
14:47:14)
- https://wiki.acumos.org/display/AR/Meetings
(farheen,
14:47:44)
- recording of demo recorded in link
above. (farheen,
14:48:04)
- the models with high bias their bias is removed
from the nifi data. (farheen,
14:51:22)
- will Ai360 be a library or service? Will it be
deployed on need to know basis? (farheen,
14:51:51)
- all 3 possible. pipinstall ai360 python
library. Can be wrapped in any framework. Also can be provided as
rest. (farheen,
14:52:44)
- ACTION: Nat form a
focus team and call around Ai 360. Once determined we can propose
to TSC. (farheen,
14:53:14)
- Nat - The Security verification and ml
workbench teams. (farheen,
14:53:39)
- Sayee - Model training steps? (farheen,
14:53:57)
- - Romeo - Yes, it can be integrated or as a
separate microservice. It is up to the team to decide. (farheen,
14:54:31)
- Nat - We have to have a focused discussion.
There are two other products Robust and inaudible. (farheen,
14:55:18)
- Romeo - I will be in Lyon next week.
(farheen,
14:56:05)
- Nat and Phillippe will be there. Martin will
be there too. (farheen,
14:56:43)
- ACTION: Nat report
out on your findings at the face to face in Lyon. (farheen,
14:57:28)
- ACTION: Murali add
the Demeter release events to the lists.acumos.org calendar.
(farheen,
15:00:29)
Meeting ended at 15:00:36 UTC
(full logs).
Action items
- Guy create a task on Epic 3256 regarding meeting with Tommy.
- Manoop add Vineet's demo of logs in Clio to next weeks agenda.
- Nat form a focus team and call around Ai 360. Once determined we can propose to TSC.
- Nat report out on your findings at the face to face in Lyon.
- Murali add the Demeter release events to the lists.acumos.org calendar.
People present (lines said)
- farheen (59)
- collabot (3)
Generated by MeetBot 0.1.4.