Add your own splitting strategy

Add your own splitting strategy#

Even though molflux.splits comes with built-in splitting strategies, you may want to add your own splitter. In this guide, we go through two ways you can do this.

Temporary splitting strategy#

The first method is by registering your strategy temporarily. This is useful in cases when you are still prototyping and would like to have easy access to your strategy code and modify it.

For a splitting strategy to be available in molflux.splits, it must have a splitting strategy class. The basic skeleton looks like this

from typing import Any, Iterator, Optional

from molflux.splits.catalogue import register_splitting_strategy
from molflux.splits.bases import SplittingStrategyBase
from molflux.splits.info import SplittingStrategyInfo
from molflux.splits.typing import ArrayLike, SplitIndices, Splittable

_DESCRIPTION = """
My new splitting strategy
"""


@register_splitting_strategy(kind = "custom", name = "your_split_name")
class YourNewSplit(SplittingStrategyBase):
    def _info(self) -> SplittingStrategyInfo:
        return SplittingStrategyInfo(
            description=_DESCRIPTION
        )

    def _split(
        self,
        dataset: Splittable,
        y: Optional[ArrayLike] = None,
        groups: Optional[ArrayLike] = None,
        **kwargs: Any,
    ) -> Iterator[SplitIndices]:


        # my new splitting strategy
        train_indices = [0, 3, 5]
        validation_indices = [1, 2]
        test_indices = [4]

        yield train_indices, validation_indices, test_indices

from molflux.splits import load_splitting_strategy

strategy = load_splitting_strategy(name='your_split_name')

print(strategy)

SplittingStrategy(
	name: "your_new_split",
	tag: "your_new_split",
	signature: self.split(dataset: collections.abc.Sized, y: Optional[collections.abc.Iterable] = None, groups: Optional[collections.abc.Iterable] = None, **kwargs: Any) -> Iterator[tuple[collections.abc.Iterable[int], collections.abc.Iterable[int], collections.abc.Iterable[int]]],
	description: """
My new splitting strategy
""",
	usage: """"""
	state: {}
)

Let’s break this down.

Start by creating a class (you can name it whatever you like, the class name has no impact on how the strategy will appear in molflux.splits). This class must inherit SplittingStrategyBase from molflux.splits.

Next, since we are adding this splitting strategy temporarily, you need to register it (basically make it available in molflux.splits) by adding the decorator register_splitting_strategy to the class. Here, you specify the kind and name of your strategy. This is how it will appear in the molflux.splits catalogue and name is how it will be loaded.

Next, you should also add a description to your strategy. This is done by adding an _info method which returns a SplittingStrategyInfo object.

Finally, you need to implement the _split method. This is the method used to compute the split indices. There are two requirements:

  • Signature: The _split method must have the following arguments

    • A dataset argument of type Splittable. This is the data to be split (Splittable enforces that it has a size).

    • y, the target variable for supervised learning problems. Default to None if not needed.

    • group, group labels for the samples used while splitting the dataset. Default to None if not needed

    • After that, you can add all the optional arguments for splitting with default values.

    • Finally, you need to have **kwargs: Any (this is for mypy type checking reasons).

  • Return type: The _split method must yield a tuple of (train_indices, validation_indices, test_indices).

Et voilà ! As you can see from the python cell above, the strategy was loaded using molflux.splits!

Permanent splitting strategy#

If you have settled on a splitting strategy and would like to have it as part of a package, you can still use the above method to do this (add a script to your repo with the above code). If you do this though, your strategy will only be available in molflux.splits if you have imported your package (and also imported the strategy class from your package).

To avoid this, we strongly recommended that you add your strategy as a plugin (to learn more about plugins see here). Using this method, your strategy will appear and be available in molflux.splits automatically (you need to have it pip installed in the same environment).

To begin with, add your strategy class to your repo (this can be any repo, not necessarily molflux). Next, remove the register_splitting_strategy decorator (it’s no longer needed).

from typing import Any, Iterator, Optional

from molflux.splits.bases import SplittingStrategyBase
from molflux.splits.info import SplittingStrategyInfo
from molflux.splits.typing import ArrayLike, SplitIndices, Splittable

_DESCRIPTION = """
My new splitting strategy
"""

class YourNewSplit(SplittingStrategyBase):
    def _info(self) -> SplittingStrategyInfo:
        return SplittingStrategyInfo(
            description=_DESCRIPTION
        )

    def _split(
        self,
        dataset: Splittable,
        y: Optional[ArrayLike] = None,
        groups: Optional[ArrayLike] = None,
        **kwargs: Any,
    ) -> Iterator[SplitIndices]:


        # my new splitting strategy
        train_indices = [0, 3, 5]
        validation_indices = [1, 2]
        test_indices = [4]

        yield train_indices, validation_indices, test_indices

Now, go to the pyproject.toml file of your repo and under [project.entry-points.'molflux.splits.plugins.pluging_name'], add a plugin to your strategy class as follows

[project.entry-points.'molflux.splits.plugins.plugin_name']
name_of_strategy = 'path.to.module.file:YourNewSplit'

Note

You can also do this in the setup.cfg file of your repo and under [options.entry_points]. Add a plugin to your strategy class as follows

[options.entry_points]
molflux.splits.plugins.kind_of_strategy =
   name_of_strategy = path.to.module.file:YourNewSplit

Let’s break this down. This entry point allows molflux.splits to detect your strategy (you need to have both molflux and your package installed in the same environment, the order of installation does not matter). In the plugin definition, you can specify the kind of your strategy. Next, you specify the path to the class of your strategy. Here you define the name of your strategy.

And that’s it! You can now find your strategy available in molflux.splits.

Contents