TruLens Streamlit components

[sfc-gh-jreini]

Items to add to release announcement:

• Add TruLens components that can be added to streamlit dashboards

https://github.com/truera/trulens/assets/171636836/3a5a0b53-27ed-442d-9139-6f118d332400

[sfc-gh-dhuang]

FYI you can open the PR as a Draft or still click Convert to draft near the reviewer section even after PR is opened :)

[sfc-gh-jreini]

One overall thing that confuses me is that is there a reason we're branding this stuff as trulens_st? Like is it important that it's streamlit? Why not like ui or something that way if we change it later it's not a misnomer or something.

Yeah, these are streamlit components targeted at addressing people already building with Streamlit. This is useful because:

• Streamlit is oriented as a frontend for GenAI
• We already use streamlit as the TruLens frontend, so this is relatively easy to do. However the current TruLens front end is not malleable, and users often rebuild portions of it into their own streamlit dashboards.
• This makes it easier to integrate into the large set of streamlit apps used by Snowflake users

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[sfc-gh-jreini]

@sfc-gh-dkurokawa seems like your comments got squashed, but I think they should be addressed. I moved trulens_trace back to just calling tru - this should actually be okay because its a singleton.

[sfc-gh-dkurokawa]

@sfc-gh-dkurokawa seems like your comments got squashed, but I think they should be addressed. I moved trulens_trace back to just calling tru - this should actually be okay because its a singleton.

AFAICT, it's not technically a singleton, like if you give it a name you can have more then one. That is you can have:

tru_app1 = Tru(name="app1")
tru_app2 = Tru(name="app2)

as per the code it seems (I haven't tested it though).

[sfc-gh-jreini]

@sfc-gh-dkurokawa seems like your comments got squashed, but I think they should be addressed. I moved trulens_trace back to just calling tru - this should actually be okay because its a singleton.

AFAICT, it's not technically a singleton, like if you give it a name you can have more then one. That is you can have:

tru_app1 = Tru(name="app1")
tru_app2 = Tru(name="app2)

as per the code it seems (I haven't tested it though).

Ran the following test that seems to indicate it is a singleton/this would be okay. Let me know if you disagree

Set up two instances with different parameters and check fi they're equal:

from trulens_eval import Tru

tru_app1 = Tru(database_url='sqlite:///memory')
tru_app2 = Tru()
print(tru_app1 is tru_app2)

True

Manually check arguments from app 1.

args_tru_app1 = tru_app1.__dict__
print("Arguments used to initialize tru_app1:", args_tru_app1)

Arguments used to initialize tru_app1: {'db': SQLAlchemyDB(redact_keys=False, table_prefix='trulens_', engine_params={'url': 'sqlite:///memory',

Manually check arguments from app 2.

args_tru_app2 = tru_app2.__dict__
print("Arguments used to initialize tru_app2:", args_tru_app2)

Arguments used to initialize tru_app2: {'db': SQLAlchemyDB(redact_keys=False, table_prefix='trulens_', engine_params={'url': 'sqlite:///memory', ...

Arguments from app 2 are actually from app 1