Create a Machine Learning Workflow

Because Pachyderm is a language and framework agnostic and platform, and because it easily distributes analysis over large data sets, data scientists can use any tooling for creating machine learning workflows. Even if that tooling is not familiar to the rest of an engineering organization, data scientists can autonomously develop and deploy scalable solutions by using containers. Moreover, Pachyderm’s pipeline logic paired with data versioning make any results reproducible for debugging purposes or during the development of improvements to a model.

For maximum leverage of Pachyderm’s built functionality, Pachyderm recommends that you combine model training processes, persisted models, and model utilization processes, such as making inferences or generating results, into a single Pachyderm pipeline Directed Acyclic Graph (DAG).

Such a pipeline enables you to achieve the following goals:

  • Keep a rigorous historical record of which models were used on what data to produce which results.
  • Automatically update online ML models when training data or parameterization changes.
  • Easily revert to other versions of an ML model when a new model does not produce an expected result or when bad data is introduced into a training data set.

The following diagram demonstrates an ML pipeline:

Example of a machine learning workflow

You can update the training dataset at any time to automatically train a new persisted model. Also, you can use any language or framework, including Apache Spark™, Tensorflow™, scikit-learn™, or other, and output any format of persisted model, such as pickle, XML, POJO, or other. Regardless of the framework, Pachyderm versions the model so that you can track the data that was used to train each model.

Pachyderm processes new data coming into the input repository with the updated model. Also, you can recompute old predictions with the updated model, or test new models on previously input and versioned data. This feature enables you to avoid manual updates to historical results or swapping ML models in production.

For examples of ML workflows in Pachyderm see Machine Learning Examples.