### Scaling with Out-of-core Ops¶

• Used when data cannot fit in main memory. You can stream instances, extract features from instances, or use an incremental algorithm.

• Incremental learning options are supported in Scikit-Learn via any estimator that has a partial_fit parameter.

• Incremental classifiers may be unable to cope with new, unseen target classes. In this case you'll have provide all possible classes to the first call to partial_fit through classes'.

• Be aware that some algorithms do not weigh samples equally over time. Perceptron` is an example of this.

### Example: OOC text document classification¶

• Demo of OOC learning with batches of example data.
• Use HashingVectorizer to project each example into the same feature space, so that it remains the same over time - especially useful when new words may appear in each batch.