Streamlit webapp .get_predictor()

Matteo

Hi everyone,

I am attempting to convert a Dash webapp to Streamlit. The app simply takes inputs parameters from the users and generates predictions based on a model trained in the same flow where the app is stored.

Although everything works smoothly in both Dash and Streamlit, I am encountering issues when attempting to call the machine learning models in Streamlit.

Below is the code snippet that is returning an error in Streamlit:

model = dataiku.Model("3cxcmJox")
model_pred = model.get_predictor()

FileNotFoundError: [Errno 2] No such file or directory: '/home/dataiku/data/saved_models/FLOW/3cxcmJox/versions/1664296109477/core_params.json'

File "/opt/dataiku/python-code-envs/pyenv-streamlit/lib/python3.7/site-packages/streamlit/scriptrunner/script_runner.py", line 475, in _run_script
exec(code, module.__dict__)

File "/home/dataiku/workspace/code_studio-versioned/streamlit/app.py", line 122, in <module>
model.get_predictor()

File "/opt/dataiku/python/dataiku/core/saved_model.py", line 278, in get_predictor
self._predictors[version_id] = build_predictor_for_saved_model(model_folder, sm.get("conditionalOutputs", []))

File "/opt/dataiku/python/dataiku/core/saved_model.py", line 402, in build_predictor_for_saved_model
core_params = dkujson.load_from_filepath(osp.join(model_folder, "core_params.json"))

File "/opt/dataiku/python/dataiku/core/dkujson.py", line 37, in load_from_filepath
with open(filepath) as fp:

Does anyone have a solution to this issue?

Thanks in advance

Alexandru

Hi @mfnz
,
The get_predictor does not currently work in a container and streamlit webapp runs only in containerized execution.
For now, you won't be able to use get_predictor in the streamlet webapp as this now requires direct access to the DSS filesystem.

We are working on allowing get_predictor to in a containerized environment in a future release.

Thanks

sreejay222

Hi AlexT,

is there any other way we can read the saved model to containers?

importthepandas

this is quite limiting actually, for apps and api based workflows. looking forward to this being updated.

Alexandru

@mfnz
@importthepandas
@sreejay222


I just wanted to let you know that get_predictor can now work in containerized environment starting with DSS 12.3.0

https://doc.dataiku.com/dss/latest/release_notes/12.html#machine-learning

• Added ability to use a Visual model’s predictor Python API from code running in containerized execution