MLFlow error

[gar]

I am running into an error within the Databricks notebook (on Databricks website) environment where MLFlow will not load:

MLflow autologging encountered a warning: "/databricks/python/lib/python3.8/site-packages/mlflow/utils/autologging_utils/safety.py:216: UserWarning: Logging to MLflow failed: You haven't configured the CLI yet! Please configure by entering

/databricks/python_shell/scripts/PythonShell.py configure

I then try to run the command in a cell (and in the web terminal) :

!/databricks/python_shell/scripts/PythonShell.py configure

And get the following error output:

/databricks/python_shell/scripts/PythonShell.py: line 22: syntax error near unexpected token `'AGG''

I have tried editing this file from the Web Terminal with vi but nothing seems to work.

[sean_owen]

How are you running this? This generally means you're not running a Databricks ML runtime, which should be all set up with auth, etc

[Anonymous]

Hi,

I am having the same issue when trying to run an MLFlor experiment with :

mlflow run git-uri -b databricks --backend-config cluster-spec.json experiment-name mlflow_expetiment

The error is caused by:

mlflow.set_tracking_uri('databricks')

The error is the following:

databricks_cli.utils.InvalidConfigurationError: You haven't configured the CLI yet! Please configure by entering `/opt/project/mlflow_test.py configure`

The databricks-cli is configured properly and works fine outside the script.

[sean_owen]

Have you tried running that script it indicates? something is not quite connected here

[Anonymous]

Hi,

Yes​ I have tried but I get the error that the script doesn't have a "configure" argument. Looking deeper into the code that reises the error, I see that this is a string automatically generated by inserting the name of the running script when the Databricks credentials are not properly fetch within python. But I don't know how I can pass this credentials that I have already pass to the databricks-cli (token, host, etc)

[erocoar]

I get the error when I run `!pip install mlflow` instead of `%pip install mlflow`, which I assume installs the databricks-maintained version

[sean_owen · Accepted solution]

There aren't different versions of mlflow, but without %pip install you are only installing on the driver machine. You do need %pip to even get it on the workers, which could be the issue.

[sj2812]

step1 - !pip install mlfow , step2- mlflow.set_tracking_uri('databricks'), step3- restart the cluster

[sean_owen]

Do not use !pip. This installs on the driver only. Use %pip. However, you do not need to install MLflow. Use the ML runtime. Otherwise you also have to configure MLflow.

[Will1]

Install the Databricks SQL Connector for Python library on your development machine by running pip install databricks-sql-connector .

1. Query data.
2. Insert data.
3. Query metadata.
4. Cursor and connection management.
5. Configure logging.

Regards,

Willjoe