mlflow cli not linked in path after pip install

[nkarpovdb]

System information

• Have I written custom code (as opposed to using a stock example script provided in MLflow): no
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu, 18.04 LTS (amazon ami)
• MLflow installed from (source or binary): binary
• MLflow version (run mlflow --version): n/a
• Python version: 2.7.14
• Exact command to reproduce: pip install mlflow, mlflow ui

Describe the problem

Clean install doesn't link mlflow to any of the directories usually on the path (/usr/local/sbin:/usr/local/bin etc)

The install does work otherwise (... python2.7/site-packages/mlflow/cli.py exists)

Source code / logs

Successfully built mleap querystring-parser simplejson pyyaml itsdangerous nose-exclude tabulate configparser MarkupSafe
Installing collected packages: numpy, six, setuptools, protobuf, jmespath, docutils, python-dateutil, urllib3, botocore, futures, s3transfer, boto3, Werkzeug, click, MarkupSafe, Jinja2, itsdangerous, Flask, gunicorn, smmap2, gitdb2, gitpython, pytz, pandas, scipy, scikit-learn, argparse, nose, nose-exclude, mleap, idna, certifi, chardet, requests, querystring-parser, simplejson, pyyaml, tabulate, configparser, databricks-cli, mlflow

ubuntu@ip-###:~$ mlflow ui

Command 'mlflow' not found, did you mean:

  command 'mflow' from deb mblaze

Try: sudo apt install <deb name>

[mparkhe]

Hi @nkarpovdb : Are you using some virtual env before installing mlflow? Either of virtualenv and conda create before doing a pip install should help. Please lemme know if that does not help.

[nenetto]

I usually use virtualenvand sometimes, for python packages with scripts that need to be added to path are not correctly found by the OS due to the virtual environment. To fix this, I have to install the package in the root python installation without activating any environment so script can be found properly.

Try to install mlflow in the root python installation and deactivating first the environment. Then, install mlflow in the environment you want to use mlflow from. That should fix your issue

[stevenkwong]

I am using anaconda, and run in root env. It practical for me, you can have a try:
after pip mlflow, you can find the exec file in the fellowing path:

${anaconda path}/bin/mlflow

then build a soft link, and you can use "mlflow" in command line.

[conradbm]

Where does mlflow drop the application mlflow that can leverage the command mlflow ui? I am trying to find it, but I am not clear on where it is in the folder structure.

[TFrascaroli]

Where does mlflow drop the application mlflow that can leverage the command mlflow ui? I am trying to find it, but I am not clear on where it is in the folder structure.

@conradbm After installing it, it told me where it installed to, warning me that its path was not in PATH. For me (and I guess for everyone) it is located at /home/<username>/.local/bin. Then doing
PATH=$PATH:/home/<username>/.local/bin
fixed the issue, and now I can run mlflow ui.