def test_explain_model_pandas_string(self, tabular_explainer):
np.random.seed(777)
num_rows = 100
num_ints = 10
num_cols = 4
split_ratio = 0.2
A = np.random.randint(num_ints, size=num_rows)
B = np.random.random(size=num_rows)
C = np.random.randn(num_rows)
cat = np.random.choice(['New York', 'San Francisco', 'Los Angeles',
'Atlanta', 'Denver', 'Chicago', 'Miami', 'DC', 'Boston'], 100)
label = np.random.choice([0, 1], num_rows)
df = pd.DataFrame(data={'A': A, 'B': B, 'C': C, 'cat': cat, 'label': label})
df.cat = df.cat.astype('category')
X = df.drop('label', axis=1)
y = df.label
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=split_ratio)
clf = create_lightgbm_classifier(x_train, y_train)
explainer = tabular_explainer(clf, initialization_examples=x_train, features=x_train.columns)
global_explanation = explainer.explain_global(x_test)
local_shape = global_explanation._local_importance_values.shape
num_rows_expected = split_ratio * num_rows
assert local_shape == (2, num_rows_expected, num_cols)
assert len(global_explanation.global_importance_values) == num_cols
assert global_explanation.num_features == num_cols
def test_raw_timestamp_explanation(self, mimic_explainer):
df = retrieve_dataset(
'insurance_claims.csv',
na_values='?',
parse_dates=['policy_bind_date', 'incident_date'])
label = 'fraud_reported'
df_y = df[label]
df_X = df.drop(columns=label)
x_train, x_test, y_train, y_test = train_test_split(df_X,
df_y,
test_size=0.2,
random_state=7)
str_cols = df_X.select_dtypes(
exclude=[np.number, np.datetime64]).columns.tolist()
dt_cols = df_X.select_dtypes(include=[np.datetime64]).columns.tolist()
numeric_cols = df_X.select_dtypes(include=[np.number]).columns.tolist()
transforms_list = []
for str_col in str_cols:
transforms_list.append(
(str_col,
Pipeline(steps=[('imputer',
SimpleImputer(strategy='most_frequent')
), ('ohe', OneHotEncoder(sparse=False))]),
[str_col]))
for numeric_col in numeric_cols:
transforms_list.append(
(numeric_col,
Pipeline(steps=[('imputer', SimpleImputer(
strategy='mean')), ('scaler', StandardScaler())]),
[numeric_col]))
for dt_col in dt_cols:
transforms_list.append(
(dt_col, Pipeline(steps=[('scaler', StandardScaler())]),
[dt_col]))
transformations = ColumnTransformer(transforms_list)
x_train_transformed = transformations.fit_transform(x_train)
model = create_lightgbm_classifier(x_train_transformed, y_train)
model_task = ModelTask.Classification
features = df_X.columns.tolist()
explainer = mimic_explainer(model,
x_train,
LGBMExplainableModel,
transformations=transformations,
features=features,
model_task=model_task)
explanation = explainer.explain_global(x_train)
dashboard_pipeline = Pipeline(steps=[('preprocess',
transformations), ('model',
model)])
ExplanationDashboard(explanation,
dashboard_pipeline,
datasetX=x_train,
trueY=y_train)
def test_explain_model_lightgbm_multiclass(self, tabular_explainer, iris):
# Fit a lightgbm model
model = create_lightgbm_classifier(iris[DatasetConstants.X_TRAIN], iris[DatasetConstants.Y_TRAIN])
# Create tabular explainer
exp = tabular_explainer(model, iris[DatasetConstants.X_TRAIN], features=iris[DatasetConstants.FEATURES],
classes=iris[DatasetConstants.CLASSES])
test_logger.info('Running explain global for test_explain_model_lightgbm_multiclass')
explanation = exp.explain_global(iris[DatasetConstants.X_TEST])
assert len(explanation.local_importance_values[0]) == len(iris[DatasetConstants.X_TEST])
assert explanation.num_examples == len(iris[DatasetConstants.X_TEST])
assert len(explanation.local_importance_values) == len(iris[DatasetConstants.CLASSES])
assert explanation.num_classes == len(iris[DatasetConstants.CLASSES])
def test_explain_model_lightgbm_binary(self, tabular_explainer, iris):
X, y = shap.datasets.adult()
x_train, x_test, y_train, _ = train_test_split(X,
y,
test_size=0.2,
random_state=7)
# Fit a tree model
model = create_lightgbm_classifier(x_train, y_train)
classes = ["<50k", ">50k"]
# Create local tabular explainer without run history
exp = tabular_explainer(model,
x_train,
features=X.columns.values,
classes=classes)
test_logger.info(
'Running explain global for test_explain_model_lightgbm_binary')
explanation = exp.explain_global(x_test)
assert len(explanation.local_importance_values[0]) == len(x_test)
assert len(explanation.local_importance_values) == len(classes)
def create_model(x_train, y_train):
return create_lightgbm_classifier(x_train, y_train)