Creating a Custom Check

It is possible to extend deepchecks by implementing custom checks. This enables you to have your own logic of metrics or validation, or even just to display your own graph using deepchecks’ suite.

• Check Structure

• Write a Basic Check

• Check Display

• Custom Check Templates

Check Structure

Each check consists of 3 main parts:

• Return Value

• Display

• Conditions

This guide will demonstrate how to implement a Check with a return value and display, for adding a condition see Configure Check Conditions, or have a look at the examples in Custom Check Templates guide..

Write a Basic Check

Let’s implement a check for comparing the sizes of the test and the train datasets.

The first step is to create check class, which inherits from a base check class. Each base check is differed by its run method signature, read more about all types. In this case we will use TrainTestBaseCheck, which is used to compare between the test and the train datasets. After creating the basic class with the run_logic function we will write our check logic inside it.

Good to know: the return value of a check can be any object, a number, dictionary, string, etc…

The Context Object

The logic of all tabular checks is executed inside the run_logic() function. The sole argument of the function is the context object, which has the following optional members:

• train: the train dataset

• test: the test dataset

• model: the model

When writing your run_logic() function, you can access the train and test datasets using the context object. For more examples of using the Context object for different types of base checks, see the Custom Check Templates guide.

Check Example

from deepchecks.core import CheckResult
from deepchecks.tabular import Context, Dataset, TrainTestCheck


class DatasetSizeComparison(TrainTestCheck):
    """Check which compares the sizes of train and test datasets."""

    def run_logic(self, context: Context) -> CheckResult:
        ## Check logic
        train_size = context.train.n_samples
        test_size = context.test.n_samples

        ## Return value as check result
        return_value = {'train_size': train_size, 'test_size': test_size}
        return CheckResult(return_value)

Hooray! we just implemented a custom check. Now let’s create two Datasets and try to run it:

import pandas as pd

# We'll use dummy data for the purpose of this demonstration
train_dataset = Dataset(pd.DataFrame(data={'x': [1,2,3,4,5,6,7,8,9]}), label=None)
test_dataset = Dataset(pd.DataFrame(data={'x': [1,2,3]}), label=None)

result = DatasetSizeComparison().run(train_dataset, test_dataset)
result

Our check ran successfully but we got the print “Nothing found”. This is because we haven’t defined to the check anything to display, so the default behavior is to print “Nothing found”. In order to access the value that we have defined earlier we can use the “value” property on the result.

result.value

{'train_size': 9, 'test_size': 3}

To see code references for more complex checks (that can receive parameters etc.), check out any of your favorite checks from our API Reference.

Check Display

Most of the times we will want our checks to have a visual display that will quickly summarize the check result. We can pass objects for display to the CheckResult. Objects for display should be of type: html string, dataframe or a function that plots a graph. Let’s define a graph that will be displayed using matplotlib. In order to use matplotlib we have to implement the code inside a function and not call it directly in the check, this is due to architectural limitations of matplotlib.

Good to know: ``display`` can receive a single object to display or a list of objects

import matplotlib.pyplot as plt

from deepchecks.core import CheckResult
from deepchecks.tabular import Context, Dataset, TrainTestCheck


class DatasetSizeComparison(TrainTestCheck):
    """Check which compares the sizes of train and test datasets."""

    def run_logic(self, context: Context) -> CheckResult:
        ## Check logic
        train_size = context.train.n_samples
        test_size = context.test.n_samples

        ## Create the check result value
        sizes = {'Train': train_size, 'Test': test_size}
        sizes_df_for_display =  pd.DataFrame(sizes, index=['Size'])

        ## Display function of matplotlib graph:
        def graph_display():
            plt.bar(sizes.keys(), sizes.values(), color='green')
            plt.xlabel("Dataset")
            plt.ylabel("Size")
            plt.title("Datasets Size Comparison")

        return CheckResult(sizes, display=[sizes_df_for_display, graph_display])

Let us check it out

DatasetSizeComparison().run(train_dataset, test_dataset)

Voila!

Now we have a check that prints a graph and has a value. We can add this check to any Suite and it will run within it.

The next possible step is to implement a condition, which will allow us to give the check result a pass / fail mark. To do so, check out the following guide.

Base Checks Types

There are a number of different BaseCheck Classes to inherit from. Each base check is differed by the objects it requires in order to run, and their sole difference is the run method’s signature.

Check	run Signature	Notes
SingleDatasetBaseCheck	run(self, dataset, model=None)	When used in a suite you can choose whether to run on the test dataset, the train dataset or on both
TrainTestBaseCheck	run(self, train_dataset, test_dataset, model=None)
ModelOnlyBaseCheck	run(self, model)
ModelComparisonCheck	run(self, List[train_dataset], List[test_dataset], List[model])

previous

Use Cases - Classifying Malicious URLs

next

Custom Check Templates