Getting predictions

Using modelkit models

The first thing you will want to do with a Model is get predictions from it. There are three ways to do so.

Let us consider this example, in which we have implemented _predict

class MyModel(Model):
    def _predict(self, item, **kwargs):
        return item, kwargs

Predictions for single items

Predictions for a single item can be obtained by calling the object, or it's predict function.

prediction = model(item) # or model.predict(item)
# returns item

This will call whichever one of _predict or _predict_batch was implemented in the Model.

Note

Although you have implemented _predict, you need to call predict (no underscore) to get predictions. The difference between the two is where modelkit's magic operates 😀

Note that keyword arguments will be passed all the way to the implemented _predict:

prediction = model(item, some_kwarg=10) # or model.predict(item)
# returns item, {"some_kwargs": 10}

Predictions for lists of items

Predictions for list of items can also easily be obtained by using predict_batch:

items = [1,2,3,4]
predictions = model.predict_batch(items)

Here, items must be a list of items. modelkit will iterate on it and fetch predictions.

This will also call whichever one of _predict or _predict_batch was implemented in the Model, and pass kwargs.

Predictions from iterators

It is also possible to iterate through predictions with an iterator, which is convenient to avoid having to load all items to memory before getting predictions.

def generate_items():
    ...
    yield item
for prediction in model.predict_gen(generate_items()):
     # use prediction
    ...

A typical use case is to iterate through the lines of a file, perform some processing and write it straight back to another file

Note that in the case in which _predict_batch is implemented, you may see speed ups if you have written vectorized code. See the documentation on batching for more information