Managing the Machine Learning Lifecycle with MLflow
4 likes1,071 views
The document presents an agenda for a presentation on MLflow, highlighting its role in the deployment lifecycle of machine learning models. It discusses the complexities of deploying ML models compared to traditional software, emphasizing data management, model training, and monitoring. MLflow is introduced as an open-source platform that provides features for tracking experiments, managing lifecycle, and improving collaboration in model development.
![X, y = get_training_data()
opt = keras.optimizers.Adam(lr=params["learning_rate"],
beta_1=params["beta_1"],
beta_2=params["beta_2"],
epsilon=params["epsilon"])
model = Sequential()
model.add(Dense(int(params["units"]), ...)
model.add(Dense(1))
model.compile(loss="mse", optimizer=opt)
rest = model.fit(X, y, epochs=50, batch_size=64, validation_split=.2)](https://image.slidesharecdn.com/2020-01-16-managingthemachinelearninglifecyclewithmlflowmlopsparis-200128145349/85/Managing-the-Machine-Learning-Lifecycle-with-MLflow-23-320.jpg)
![# Enable MLflow Autologging
mlflow.keras.autolog()
X, y = get_training_data()
opt = keras.optimizers.Adam(lr=params["learning_rate"],
beta_1=params["beta_1"],
beta_2=params["beta_2"],
epsilon=params["epsilon"])
model = Sequential()
model.add(Dense(int(params["units"]), ...)
model.add(Dense(1))
model.compile(loss="mse", optimizer=opt)
rest = model.fit(X, y, epochs=50, batch_size=64, validation_split=.2)](https://image.slidesharecdn.com/2020-01-16-managingthemachinelearninglifecyclewithmlflowmlopsparis-200128145349/85/Managing-the-Machine-Learning-Lifecycle-with-MLflow-24-320.jpg)
More Related Content
Similar to Managing the Machine Learning Lifecycle with MLflow (20)
Managing the Machine Learning Lifecycle with MLflow
- 1. Arduino Cascella @ MLOps Paris, January 16th 2020
- 2. Agenda • What is MLflow? • Demo • Questions & Answers
- 3. Deployment lifecycle of ML Models?
- 4. Can you find what’s wrong with this true story?
- 5. Can you find what’s wrong with this true story? Did anything change in the feature engineering?
- 6. Can you find what’s wrong with this true story? How did the hyperparameters change?
- 7. Can you find what’s wrong with this true story? What data was this model trained on?
- 8. Can you find what’s wrong with this true story? How did the offline metrics change?
- 9. Can you find what’s wrong with this true story? ....
- 10. The difference between releasing Software and deploying ML Models
- 11. Write code Software Write unit tests Send for review Get approvals Commit Release testing Release
- 12. Write code Software ML Models Write unit tests Send for review Get approvals Commit Release testing Release Analyze data Put data into the right format Write model code Train and evaluate model Experiment with params, model structure Deploy … by email? Monitor performance and trigger retraining
- 13. Write code Software ML Models Write unit tests Send for review Get approvals Commit Release testing Release Analyze data Put data into the right format Write model code Train and evaluate model Experiment with params, model structure Deploy … by email? Monitor performance and trigger retraining Meet a functional specification Optimize a metric, e.g. CTR Goal
- 14. Write code Software ML Models Write unit tests Send for review Get approvals Commit Release testing Release Analyze data Put data into the right format Write model code Train and evaluate model Experiment with params, model structure Deploy … by email? Monitor performance and trigger retraining Meet a functional specification Optimize a metric, e.g. CTR Goal Depends on code Depends on data, code, model, params, ... Quality
- 15. Write code Software ML Models Write unit tests Send for review Get approvals Commit Release testing Release Analyze data Put data into the right format Write model code Train and evaluate model Experiment with params, model structure Deploy … by email? Monitor performance and trigger retraining Meet a functional specification Optimize a metric, e.g. CTR Goal Depends on code Depends on data, code, model, params, ... Quality Typically one software stack Combination of many libraries, tools, ... Tools
- 16. Write code Software ML Models Write unit tests Send for review Get approvals Commit Release testing Release Analyze data Put data into the right format Write model code Train and evaluate model Experiment with params, model structure Deploy … by email? Monitor performance and trigger retraining Meet a functional specification Optimize a metric, e.g. CTR Goal Depends on code Depends on data, code, model, params, ... Quality Typically one software stack Combination of many libraries, tools, ... Tools Works deterministically Keeps changing with data, etc. Outcome
- 17. In summary, deploying ML Models is hard!
- 18. : An Open Source ML Platform 18 mlflow.org github.com/mlflow twitter.com/MLflowdatabricks.com/mlflow
- 19. 140 120 100 80 60 40 20 0 0 5 10 15 20 25 30 35 40 45 Months since Project Launch #ofContributors
- 20. Components Tracking Record and query experiments: code, data, config, results Projects Packaging format for reproducible runs on any platform Models General format that standardizes deployment paths Model Registry Centralized and collaborative model lifecycle management new 20 mlflow.org github.com/mlflow twitter.com/MLflowdatabricks.com/mlflow
- 21. Tracking Notebooks Local Apps Cloud Jobs UI API Tracking Server Parameters Metrics Artifacts ModelsMetadata Spark Data Source
- 22. Tracking Record and query experiments: code, configs, results, …etc import mlflow # log model’s tuning parameters with mlflow.start_run(): mlflow.log_param("layers", layers) mlflow.log_param("alpha", alpha) # log model’s metrics mlflow.log_metric("mse", model.mse()) mlflow.log_artifact("plot", model.plot(test_df)) Tracking API
- 23. X, y = get_training_data() opt = keras.optimizers.Adam(lr=params["learning_rate"], beta_1=params["beta_1"], beta_2=params["beta_2"], epsilon=params["epsilon"]) model = Sequential() model.add(Dense(int(params["units"]), ...) model.add(Dense(1)) model.compile(loss="mse", optimizer=opt) rest = model.fit(X, y, epochs=50, batch_size=64, validation_split=.2)
- 24. # Enable MLflow Autologging mlflow.keras.autolog() X, y = get_training_data() opt = keras.optimizers.Adam(lr=params["learning_rate"], beta_1=params["beta_1"], beta_2=params["beta_2"], epsilon=params["epsilon"]) model = Sequential() model.add(Dense(int(params["units"]), ...) model.add(Dense(1)) model.compile(loss="mse", optimizer=opt) rest = model.fit(X, y, epochs=50, batch_size=64, validation_split=.2)
- 28. Projects Project Spec Code MetadataConfig Local Execution Remote Execution
- 29. my_project/ ├── MLproject │ │ │ │ │ ├── conda.yaml ├── main.py └── model.py ... conda_env: conda.yaml entry_points: main: parameters: training_data: path lambda: {type: float, default: 0.1} command: python main.py {training_data} {lambda} $ mlflow run git://<my_project> mlflow.run(“git://<my_project>”, ...) Projects
- 30. Model Format Flavor 2Flavor 1 Simple model flavors usable by many tools Containers Batch & Stream Scoring Cloud Inference Services In-Line Code Models
- 31. What’s deployed in production? Who did it? How can I rollback to the previous model? Can I get approval before deploying? Where do I store and archive my models? Registry motivation my_model my_model_last my_model_new
- 32. Automated Jobs REST Serving Downstream Users Reviewers + CI/CD Tools Model Registry Experimental Staging A/B Tests Production Model Registry Data Scientists Deployment Engineers
- 33. MLflow Model Registry Benefits One Collaborative Hub: The Model Registry provides a central hub for making models discoverable, improving collaboration and knowledge sharing across the organization. Manage the entire Model Lifecycle (MLOps): The Model Registry provides lifecycle management for models from experimentation to deployment, improving reliability and robustness of the model deployment process. Visibility and Governance: The Model Registry provides full visibility into the deployment stage of all models, who requested and approved changes, allowing for full governance and auditability.
- 34. MLflow Model Registry Benefits One Collaborative Hub: The Model Registry provides a central hub for making models discoverable, improving collaboration and knowledge sharing across the organization. Manage the entire Model Lifecycle (MLOps): The Model Registry provides lifecycle management for models from experimentation to deployment, improving reliability and robustness of the model deployment process. Visibility and Governance: The Model Registry provides full visibility into the deployment stage of all models, who requested and approved changes, allowing for full governance and auditability.
- 35. MLflow Model Registry Benefits One Collaborative Hub: The Model Registry provides a central hub for making models discoverable, improving collaboration and knowledge sharing across the organization. Manage the entire Model Lifecycle (MLOps): The Model Registry provides lifecycle management for models from experimentation to deployment, improving reliability and robustness of the model deployment process. Visibility and Governance: The Model Registry provides full visibility into the deployment stage of all models, who requested and approved changes, allowing for full governance and auditability.
- 36. MLflow Model Registry Benefits One Collaborative Hub: The Model Registry provides a central hub for making models discoverable, improving collaboration and knowledge sharing across the organization. Manage the entire Model Lifecycle (MLOps): The Model Registry provides lifecycle management for models from experimentation to deployment, improving reliability and robustness of the model deployment process. Visibility and Governance: The Model Registry provides full visibility into the deployment stage of all models, who requested and approved changes, allowing for full governance and auditability.
- 37. Using the latest production Model
- 39. Deploying the new best Model as a new version to the Model Registry
- 42. Using the new production Model
- 44. Model Deployment Options Containers Batch & Stream Scoring Cloud Inference Services In-Line Code OSS Inference Solutions
- 45. pip install mlflow 45 mlflow.org github.com/mlflow twitter.com/MLflowdatabricks.com/mlflow
- 46. pip install mlflow 46 mlflow.org github.com/mlflow twitter.com/MLflowdatabricks.com/mlflow Try Managed MLflow on Databricks Community Edition! databricks.com/try