Skip to content

Latest commit

 

History

History
124 lines (71 loc) · 3.46 KB

README.md

File metadata and controls

124 lines (71 loc) · 3.46 KB
Raw

MLflow-Basics

How to create projects in MLflow

Step 1: Install MLflow by pip install mlflow.

Step 2: Create a conda.yaml file to give all the requirements of the environment on which you want to work. For example

name: nameofenvironment
channels:
  - defaults
dependencies:
  - numpy=1.14.3
  - pandas=0.22.0
  - scikit-learn=0.19.1
  - pip:
    - mlflow

Step 3: Create a MLproject file in the root directory to specify your project. Here you can specify your environment, entry point and also many things for defining your project workflow. For example

name: projectname

conda_env: conda.yaml

entry_points:
  main:
    parameters:
      alpha: {type: float, default: 0.5}
      l1_ratio: {type: float, default: 0.1}
    command: "python train.py {alpha} {l1_ratio}"

That's it for creating the project. Now lets see how we can run the project and visualize logs defined by us, using MLflow UI.

  • Run mlflow run . command inside your project folder. It will create the environment specified by you and executes the steps defined by you in your MLproject file.

  • Run mlflow ui to see all the logs of parameters and hyperparameters, versions, metrics and artifacts(e.g. model.pkl).

Now your model is ready to serve. Here is how you can serve your model.

  • Run the following command. It will create the model server.
mlflow models serve -m path_to_your_saved_model(mlruns/0/experiment/artifacts/model) -p 1234
  • You can make a CURL request to get the output from your model. For example:
curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["alcohol", "chlorides", "citric acid", "density", "fixed acidity", "free sulfur dioxide", "pH", "residual sugar", "sulphates", "total sulfur dioxide", "volatile acidity"],"data":[[12.8, 0.029, 0.48, 0.98, 6.2, 29, 3.33, 1.2, 0.39, 75, 0.66]]}' http://127.0.0.1:1234/invocations

N.B. You can create multiple experiments to compare model performance by the following command:

mlflow experiments create --experiment-name="name"
  • For hyparameter tuning or multi step workflow we can always configure the MLproject file and run different experiments.

Examples:

Dockerize your model

Step 1: In the root directory create a dockerfile and specify your environment.
Step 2: Build the docker image and give a name.
Step 3: In the MLproject file define environment as follows:

name: docker-example

docker_env:
  image:  mlflow-docker-example

entry_points:
  main:
    parameters:
      alpha: float
      l1_ratio: {type: float, default: 0.1}
    command: "python train.py --alpha {alpha} --l1-ratio {l1_ratio}"

Step 4: Now you can run project using mlflow run command.

N.B When we run mlflow run . it will create a docker image which includes code and all the requirements of our environments and the image will be tagged as mlflow-docker-example-<git-version>. After the image is built MLflow executes the entry point (specified in MLproject file) within the container using docker run.

Dockerization of model gives us the flexibility of including non-Python dependencies in our environments unlike conda environment.

References

MLflow docs