Generate Explanations

In this section, we show you how to create explanations using the Python Client

Details about how to plot and store the explanations that will be presented in this section, can be found here:

Plot and Store Explanations

expai.ExpaiModelExplainer

To initialise this class, it is required to get a ModelFairness object using expai.ExpaiProject.get_model_explainer(). More info here.

expai.ExpaiModelExplainer.get_allowed_explanations

This method returns a ExpaiModelExplainer that enables easy explanation generation.

expai.ExpaiProject.get_allowed_explanations(self)

explainer.get_allowed_explanations()

All following methods will return an ExpaiExplanation object containing the plots used to represent the explanation. See more information here.

expai.ExpaiModelExplainer.explain_model

Explain a model based on a whole dataset or a subset representing some meaningful group for your operations.

expai.ExpaiModelExplainer.explain_model(self, sample_name: str = None, sample_id: str = None, subset_indexes: list = None, target_class: str = None)

Attribute	Description	Required
sample_name	Exact name of the sample that will be used to explain the model.	Yes, if sample_id is not provided
sample_id	Unique identifier of the sample that will be used to explain the model.	Yes, if sample_name is not provided
subset_indexes	List containing the indexes that should be considered for the explanation	No
target_class	Name of the class that should be explained	Yes if model is classifier

subset_indexes allow you to explain your model based only on a meaningful subgroup in your data. You can use expai.ExpaiProject.get_sample() to obtain a dataframe, filter and then use the desired indexes.

exp_model = explainer.explain_model(sample_name="Dataset", subset_indexes=subset_indexes)
exp_model.plot_all()

expai.ExpaiModelExplainer.explain_variable_effect

Explain how a single variable impacts the predictions in your model.

expai.ExpaiModelExplainer.explain_variable_effect(self, sample_name: str = None, sample_id: str = None, subset_indexes: list = None, target_class: str = None, variables: list = None, variables_type: dict = None)

Attribute	Description	Required
sample_name	Exact name of the sample that will be used to explain the model.	Yes, if sample_id is not provided
sample_id	Unique identifier of the sample that will be used to explain the model.	Yes, if sample_name is not provided
subset_indexes	List containing the indexes that should be considered for the explanation	No
target_class	Name of the class that should be explained	Yes if model is classifier
variables	List of variables that should be explained	Yes
variables_type	Dictionary where variables are keys and values will represent the type of each variable ('numerical' or 'categorical')	No

subset_indexes allow you to explain your model based only on a meaningful subgroup in your data. You can use expai.ExpaiProject.get_sample() to obtain a dataframe, filter and then use the desired indexes.

# Variables that we want to explain
variables = ['marital-status', 'capital-gain']

# Type for the variables (categorical or numerical)
variables_type = {'marital-status': 'categorical',
                 'capital-gain': 'numerical'}

exp_var = explainer.explain_variable_effect(sample_name="Dataset", variables=variables, variables_type=variables_type)

exp_var.plot_all()

expai.ExpaiModelExplainer.explain_sample

Explain how the prediction for a certain sample was made and the importance of each variable.

expai.ExpaiModelExplainer.explain_sample(self, sample_name: str = None, sample_id: str = None, index: int = None, subset_indexes: list = None, target_class: str = None)

Attribute	Description	Required
sample_name	Exact name of the sample that will be used to explain the model.	Yes, if sample_id is not provided
sample_id	Unique identifier of the sample that will be used to explain the model.	Yes, if sample_name is not provided
index	Index for the sample that you want to explain	Yes
subset_indexes	List containing the indexes that should be considered for the explanation	No
target_class	Name of the class that should be explained	Yes if model is classifier

If you use subset_indexes, make sure they contain the index you want to explain.

# Select the index to be explained. If we filtered, it must be within the filtered dataframe.
index_to_explain = 2

# Obtenemos la explicación
exp_sample = explainer.explain_sample(sample_name="Dataset", index=index_to_explain, subset_indexes=subset_indexes)

# Plot
exp_sample.plot_all()

WHAT IF

This module allows you to validate and activate business actions based on your analytical models on the fly.

expai.ExpaiModelExplainer.what_if

Explain how a variable will change the prediction for a single entry when it takes any of its possible values. Obtain a plot for each variable representing the prediction for each possible value when all remaining variables are kept the same.

expai.ExpaiModelExplainer.what_if(self, sample_name: str = None, sample_id: str = None, index: int = None, subset_indexes: list = None, target_class: str = None, variables: list = None, variables_type: dict = None)

Attribute	Description	Required
sample_name	Exact name of the sample that will be used to explain the model.	Yes, if sample_id is not provided
sample_id	Unique identifier of the sample that will be used to explain the model.	Yes, if sample_name is not provided
index	Index for the sample that you want to explain	Yes
subset_indexes	List containing the indexes that should be considered for the explanation	No
target_class	Name of the class that should be explained	Yes if model is classifier
variables	List of variables that should be explained	Yes
variables_type	Dictionary where variables are keys and values will represent the type of each variable ('numerical' or 'categorical')	No

subset_indexes allow you to explain your model based only on a meaningful subgroup in your data. You can use expai.ExpaiProject.get_sample() to obtain a dataframe, filter and then use the desired indexes.

# Select index to analyze
index_to_explain = 2

# Variables to analyze in this sample
variables = ['age', 'capital-gain']

# Type for the variables
variables_type = {'age': 'numerical',
                 'capital-gain': 'numerical'}

exp_what_if = explainer.what_if(sample_name="Dataset", index=index_to_explain, variables=variables, variables_type=variables_type, subset_indexes=subset_indexes)

# Plot
exp_what_if.plot_all()

expai.ExpaiModelExplainer.what_if_battle

With this explanation, you will be able to change the values you want for a single entry and obtain the new prediction and its explanation after the changes. Validate business decisions on the fly.

expai.ExpaiModelExplainer.what_if_battle(self, sample_name: str = None, sample_id: str = None, index: int = None, subset_indexes: list = None, target_class: str = None, replace_dict: dict = None, display_replace_dict: dict = None)

Attribute	Description	Required
sample_name	Exact name of the sample that will be used to explain the model.	Yes, if sample_id is not provided
sample_id	Unique identifier of the sample that will be used to explain the model.	Yes, if sample_name is not provided
index	Index for the sample that you want to explain	Yes
subset_indexes	List containing the indexes that should be considered for the explanation	No
target_class	Name of the class that should be explained	Yes if model is classifier
replace_dict	Dictionary where keys contain the name of the variable to be replaced and values are the desired new values.	Yes
display_replace_dict	If you are using display samples, you can also introduce the display value for the replacements.	No

subset_indexes allow you to explain your model based only on a meaningful subgroup in your data. You can use expai.ExpaiProject.get_sample() to obtain a dataframe, filter and then use the desired indexes.

# Select index to analyze
index_to_explain = 2

# Values to be replaced 
replace_dict = {'age': 35}

# Generate explanation
exp_what_if_battle = explainer.what_if_battle(sample_name="Dataset", index=index_to_explain, replace_dict=replace_dict, subset_indexes=subset_indexes)

# Plot
exp_what_if_battle.plot_all()