Introduction¶
The two most common ways to deploy data pipelines are batch and online. Ploomber supports both deployment options.
In batch, you obtain new data to make predictions and store them for later use. This process usually happens on a schedule. For example, you may develop a Machine Learning pipeline that runs every morning, predicts the probability of user churn, and stores such probabilities in a database table.
Alternatively, you may deploy a pipeline as an online service and expose your model as a REST API; users request predictions at any time by sending input data.
Pipeline composition¶
Before diving into deployment details, let’s introduce the concept of pipeline composition.
The only difference between a Machine Learning training pipeline and its serving counterpart is what happens at the beginning and the end.
At training time, we obtain historical data, generate features, and train a model:
At serving time, we obtain new data, generate features and make predictions using a trained model:
When the feature engineering process does not match, training-serving skew arises. Training-serving skew is one of the most common problems when deploying ML models. To fix it, Ploomber allows you to compose pipelines: write your feature generation once and re-use it to organize your training and serving pipelines; this ensures that the feature engineering code matches precisely.