https://github.com/codeofrahul/experiments_with_mlflow-mlops
This repo has a complete demonstration of performing experiment tracking using mlflow.
https://github.com/codeofrahul/experiments_with_mlflow-mlops
Last synced: 6 months ago
JSON representation
This repo has a complete demonstration of performing experiment tracking using mlflow.
- Host: GitHub
- URL: https://github.com/codeofrahul/experiments_with_mlflow-mlops
- Owner: CodeofRahul
- License: mit
- Created: 2024-12-30T04:43:51.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-01-04T07:17:55.000Z (about 1 year ago)
- Last Synced: 2025-03-25T00:45:26.696Z (10 months ago)
- Language: Python
- Size: 646 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Experiments_with_MLFlow-MLOPS
This repo has a complete demonstration of performing experiment tracking using mlflow.
## **Agenda**
1. Experiments overview
2. DVC & MLflow
3. MLflow for Data Science
4. Code demonstration
5. MLflow server architecture (AWS vs Dagshub)
6. Model Registry
## Some useful `env` commands:
- To create environment = `conda create -p python=3.8 -y`
- To check available envs = `conda env list`
- To check available envs = `conda info --envs`
- To activate environment = `conda activate `
- To install requirements.txt = `pip install -r requirements.txt`
- To check install packages = `pip list`
- To check detailed about package = `pip show package_name`
- To install package = `pip install package_name`
- To uninstall package = `pip uninstall package_name`
- To check install package = `package_name --version`
- To check list of package in venv = `pip freeze --local` or `pip list --local`
## Git commands
- To add all file = `git add .`
- To add any particular file = `git add `
- To commit = `git commit -m "commit message"`
- To push the code = `git push origin main`
## Documentation
- mlflow docs : https://mlflow.org/docs/1.25.1/index.html
- mlflow core-concepts : https://mlflow.org/#core-concepts
- Quickstarts : https://mlflow.org/#core-concepts
- mlflow models : https://mlflow.org/docs/1.25.1/models.html
- github : https://github.com/mlflow/mlflow
- mlflow tracking : https://mlflow.org/docs/latest/tracking.html
- For remote server only (Dagshub) : https://dagshub.com/
### if mlflow local tracking "mlflow ui" not working:
- **make sure you have the correct jinja2 and flask**
- Upgrade Flask and Jinja2 to compatible versions:
```
pip install --upgrade flask jinja2
```
- Alternatively, if you need specific versions due to compatibility:
```
pip install flask==2.2.5 jinja2==3.1.2
```
- Reinstall MLflow
```
pip uninstall mlflow
pip install mlflow
```
- Now run `mlflow ui` again
### Experiment vs Run in MLflow
In MLflow, `an experiment is a container that groups related runs together`, providing a structured way to organize and track related runs. An experiment can be thought of as a project or a task that you're working on. Within an experiment, you can have multiple runs, each representing a specific execution of a model or a workflow. A run is a single execution of a model or a workflow, and it's the basic unit of tracking in MLflow. Each run within an experiment represents an individual instance of a model training or evaluation process, capturing detailed information such as parameters, metrics, and artifacts. This structure allows for efficient tracking and comparison of different training iterations, facilitating model development and optimization.