CODEWITHSUNDEEP
modelregistrymlflowserve
user1513335
user1513335

Reputation: 91

Serve model using MLFlow from registry

I have been trying to serve a model using MLFlow to no avail. Here is what I'm doing:

Step 1: Run all data prep steps in my Jupyter notebook
Step 2: start an Anaconda command prompt and go the same directory of the notebook
Step 3: start mlflow as follows:

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./artifacts

Step 4: set tracking uri in the notebook as follows:

mlflow.set_tracking_uri('http://localhost:5000')

Step 5: run experiments in the notebook
Step 6: register the best experiment as production (in the notebook)
Step 7: start another command prompt and go the same directory of the notebook
Step 8: serve the registered model as follows:

mlflow models serve --model-uri models:/random-forest/Production -p 1234 --no-conda

At this stage I get the following error:

Model Registry features are not supported by the store with URI: 'file:///C:/localpath/mlruns'. Stores with the following URI schemes are supported: ['databricks', 'http', 'https', 'postgresql', 'mysql', 'sqlite', 'mssql'].

Though, I'm using a sqlite database (as seen in step 3). MLFlow is using it - because I can see the sqlite file size increase when I run experiments. Everything (including the UI) is working fine except serving the model. Can anyone tell me what I'm doing wrong?

Upvotes: 2

Views: 2655

Answers (1)

user1513335
user1513335

Reputation: 91

Solved: Right before executing the model serve command at step 8, you need to create a new environment variable (in Windows) as follows: go to Environment variables, click on New for System, and add the following entry:

Variable: MLFLOW_TRACKING_URI
Value: http://localhost:5000

Upvotes: 3

Related Questions