fiddler-client 2.3.0.dev2__py3-none-any.whl → 2.4.0.dev1__py3-none-any.whl

fiddler3/tests/apis/test_xai.py

@@ -0,0 +1,690 @@
+import io
+import os
+import shutil
+from collections import namedtuple
+
+import pandas as pd
+import responses
+
+from fiddler3.entities.model import Model
+from fiddler3.schemas.xai import DatasetDataSource, RowDataSource
+from fiddler3.tests.constants import (
+    BASE_TEST_DIR,
+    DATASET_NAME,
+    MODEL_ID,
+    MODEL_NAME,
+    ORG_ID,
+    ORG_NAME,
+    OUTPUT_DIR,
+    PROJECT_ID,
+    PROJECT_NAME,
+    URL,
+    USER_EMAIL,
+    USER_ID,
+    USER_NAME,
+)
+
+MODEL_SCHEMA = {
+    'columns': [
+        {
+            'bins': [
+                350.0,
+                400.0,
+                450.0,
+                500.0,
+                550.0,
+                600.0,
+                650.0,
+                700.0,
+                750.0,
+                800.0,
+                850.0,
+            ],
+            'categories': None,
+            'data_type': 'int',
+            'id': 'creditscore',
+            'max': 850,
+            'min': 350,
+            'n_dimensions': None,
+            'name': 'CreditScore',
+            'replace_with_nulls': None,
+        },
+    ],
+    'schema_version': 1,
+}
+MODEL_SPEC = {
+    'custom_features': [],
+    'decisions': ['Decisions'],
+    'inputs': [
+        'CreditScore',
+    ],
+    'metadata': [],
+    'outputs': ['probability_churned'],
+    'schema_version': 1,
+    'targets': ['Churned'],
+}
+MODEL_API_RESPONSE_200 = {
+    'data': {
+        'id': MODEL_ID,
+        'name': MODEL_NAME,
+        'project': {
+            'id': PROJECT_ID,
+            'name': PROJECT_NAME,
+        },
+        'organization': {
+            'id': ORG_ID,
+            'name': ORG_NAME,
+        },
+        'input_type': 'structured',
+        'task': 'binary_classification',
+        'task_params': {
+            'binary_classification_threshold': 0.5,
+            'target_class_order': ['Not Churned', 'Churned'],
+            'group_by': None,
+            'top_k': None,
+            'class_weights': None,
+            'weighted_ref_histograms': None,
+        },
+        'schema': MODEL_SCHEMA,
+        'spec': MODEL_SPEC,
+        'description': 'This model predicts whether customer stays or churns',
+        'event_id_col': None,
+        'event_ts_col': None,
+        'event_ts_format': None,
+        'xai_params': {
+            'custom_explain_methods': [],
+            'default_explain_method': None,
+        },
+        'artifact_status': 'no_model',
+        'artifact_files': [],
+        'created_at': '2023-11-22 16:50:57.705784',
+        'updated_at': '2023-11-22 16:50:57.705784',
+        'created_by': {
+            'id': USER_ID,
+            'full_name': USER_NAME,
+            'email': USER_EMAIL,
+        },
+        'updated_by': {
+            'id': USER_ID,
+            'full_name': USER_NAME,
+            'email': USER_EMAIL,
+        },
+        'input_cols': [
+            MODEL_SCHEMA['columns'][0],
+            MODEL_SCHEMA['columns'][0],
+        ],
+        'output_cols': [MODEL_SCHEMA['columns'][0]],
+        'target_cols': [MODEL_SCHEMA['columns'][0]],
+        'metadata_cols': [],
+        'decision_cols': [MODEL_SCHEMA['columns'][0]],
+        'is_binary_ranking_model': False,
+    },
+}
+
+EXPLAIN_RESPONSE_200 = {
+    'data': {
+        'explanation_type': 'FIDDLER_SHAP',
+        'num_permutations': 50,
+        'ci_level': 0.95,
+        'explanations': {
+            'probability_churned': {
+                'model_prediction': 0.40507614012286625,
+                'baseline_prediction': 0.19293562908483303,
+                'GEM': {
+                    'type': 'container',
+                    'contents': [
+                        {
+                            'type': 'simple',
+                            'feature-name': 'CreditScore',
+                            'value': 619,
+                            'attribution': -0.022117800470630538,
+                            'attribution-uncertainty': 0.006245936205277383,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'Geography',
+                            'value': 'France',
+                            'attribution': 0.028808698749482956,
+                            'attribution-uncertainty': 0.008490776240161464,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'Gender',
+                            'value': 'Female',
+                            'attribution': 0.059352095878178614,
+                            'attribution-uncertainty': 0.010825591198269912,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'Age',
+                            'value': 42,
+                            'attribution': 0.04663932271153231,
+                            'attribution-uncertainty': 0.026679728117630943,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'Tenure',
+                            'value': 2,
+                            'attribution': 0.012244800190215574,
+                            'attribution-uncertainty': 0.006979182051724819,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'Balance',
+                            'value': 0.0,
+                            'attribution': 0.08161626352219709,
+                            'attribution-uncertainty': 0.017052188444913206,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'NumOfProducts',
+                            'value': 1,
+                            'attribution': 0.038190793675380155,
+                            'attribution-uncertainty': 0.0582498509252494,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'HasCrCard',
+                            'value': 'Yes',
+                            'attribution': 0.00654585439939201,
+                            'attribution-uncertainty': 0.004515983491581958,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'IsActiveMember',
+                            'value': 'Yes',
+                            'attribution': -0.05122940826721414,
+                            'attribution-uncertainty': 0.013359099976724447,
+                        },
+                        {
+                            'type': 'simple',
+                            'feature-name': 'EstimatedSalary',
+                            'value': 101348.88,
+                            'attribution': 0.012089890649499193,
+                            'attribution-uncertainty': 0.004951759795660022,
+                        },
+                    ],
+                    'attribution': 0.21214051103803322,
+                },
+            },
+        },
+        'num_refs': 30,
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+FAIRNESS_RESPONSE_200 = {
+    'data': {
+        'protected_attributes': ['Gender'],
+        'label_distribution': {
+            'columns': ['Gender', 'data_label', 'count'],
+            'labels': [
+                ['Female', 'Churned', 79],
+                ['Female', 'Not Churned', 377],
+                ['Male', 'Churned', 39],
+                ['Male', 'Not Churned', 505],
+            ],
+        },
+        'model_outcomes': {
+            'columns': ['Gender', 'model_outcome', 'count'],
+            'outcomes': [
+                ['Female', 'Churned', 51],
+                ['Female', 'Not Churned', 405],
+                ['Male', 'Churned', 28],
+                ['Male', 'Not Churned', 516],
+            ],
+        },
+        'technical_metrics': {
+            'columns': [
+                'Gender',
+                'true_positive_rate',
+                'false_positive_rate',
+                'true_negative_rate',
+                'false_negative_rate',
+                'false_omission_rate',
+                'false_discovery_rate',
+            ],
+            'values': [
+                [
+                    'Female',
+                    0.569620253164557,
+                    0.015915119363395226,
+                    0.9840848806366048,
+                    0.43037974683544306,
+                    0.08395061728395062,
+                    0.11764705882352941,
+                ],
+                [
+                    'Male',
+                    0.5897435897435898,
+                    0.009900990099009901,
+                    0.9900990099009901,
+                    0.41025641025641024,
+                    0.031007751937984496,
+                    0.17857142857142858,
+                ],
+            ],
+        },
+        'fairness_metrics': {
+            'demographic_parity': {
+                'index': ['dp'],
+                'columns': ['Female', 'Male'],
+                'data': [[0.1118421052631579, 0.051470588235294115]],
+                'ratio': 0.4602076124567474,
+            },
+            'equal_opportunity': {
+                'index': ['tpr'],
+                'columns': ['Female', 'Male'],
+                'data': [[0.569620253164557, 0.5897435897435898]],
+                'ratio': 0.9658778205833792,
+            },
+            'group_benefit_equality': {
+                'index': ['group_benefit'],
+                'columns': ['Female', 'Male'],
+                'data': [[0.6455696202531646, 0.717948717948718]],
+                'ratio': 0.8991862567811935,
+            },
+            'disparate_impact': {
+                'pairs': [
+                    {
+                        'reference': ['Female'],
+                        'groups': [['Male']],
+                        'values': [0.4602076124567474],
+                    },
+                    {
+                        'reference': ['Male'],
+                        'groups': [['Female']],
+                        'values': [0.4602076124567474],
+                    },
+                ],
+                'absolute_min': {
+                    'groups': [['Female'], ['Male']],
+                    'value': 0.4602076124567474,
+                },
+            },
+        },
+        'model_task': 'BINARY_CLASSIFICATION',
+        'total_samples': 1000,
+        'valid_samples': 1000,
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+FETCH_SLICE_RESPONSE_200 = {
+    'data': {
+        'metadata': {
+            'query': "SELECT * FROM test_bank_churn.bank_churn WHERE geography='France' order by balance desc LIMIT 2",
+            'is_slice': True,
+            'columns': ['Age'],
+            'dtypes': ['int'],
+            'model': {
+                'id': '36bc3613-49d5-46b5-a0e0-a45c3aa4a9d9',
+                'name': 'bank_churn',
+            },
+            'env': {
+                'id': '86e59334-2b86-4726-a970-6900db46e437',
+                'type': 'PRE_PRODUCTION',
+                'name': 'test_bank_churn',
+            },
+        },
+        'rows': [[57], [37]],
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+MUTUAL_INFO_RESPONSE_200 = {
+    'data': {
+        'CreditScore': 0.009774993816146854,
+        'Geography': 1.0387361759590092,
+        'Gender': 0.000345600896640319,
+        'Age': 0.006747713128151839,
+        'Tenure': 0.0017159873190972647,
+        'Balance': 0.14278016623398132,
+        'NumOfProducts': 0.002155101383831244,
+        'HasCrCard': 0.00011152799557884174,
+        'IsActiveMember': 0.00026523254634630566,
+        'EstimatedSalary': 0.010603971349228138,
+        'probability_churned': 0.06073798415644517,
+        'Decisions': 2.3682864881247045e-05,
+        'Churned': 0.014017045772466472,
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+PREDICT_RESPONSE_200 = {
+    'data': {'predictions': [{'predicted_quality': 5.759617514660622}]},
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+FEATURE_IMPORTANCE_RESPONSE_200 = {
+    'data': {
+        'loss': 'pointwise_logloss',
+        'num_refs': 10000,
+        'ci_level': 0.5,
+        'mean_loss': 0.35653354054994635,
+        'mean_loss_ci': 0.09459192835065221,
+        'feature_names': [
+            'CreditScore',
+            'Geography',
+            'Gender',
+            'Age',
+            'Tenure',
+            'Balance',
+            'NumOfProducts',
+            'HasCrCard',
+            'IsActiveMember',
+            'EstimatedSalary',
+        ],
+        'mean_loss_increase_importance': [
+            0.004429064308612665,
+            0.11303687040447762,
+            0.010247568997879158,
+            0.20021169408760484,
+            0.020512200372182397,
+            0.36803439130037513,
+            0.2339702870224178,
+            0.0014671647143862244,
+            0.022758435725508515,
+            0.011785520720929346,
+        ],
+        'random_sample_ci': [
+            0.00021145873961269672,
+            0.0012444603359610608,
+            0.00022339657233601845,
+            0.0016025718376113422,
+            0.00020151632282746737,
+            0.0018154124659677064,
+            0.0021341170223544335,
+            0.00016283332959968953,
+            0.0010258365777659855,
+            0.0002324089593535596,
+        ],
+        'fixed_sample_ci': [
+            0.00018964651570488745,
+            0.0010810129244220677,
+            0.0001612062857999329,
+            0.001333218784229577,
+            0.000139368878515909,
+            0.001191577806041505,
+            0.0014415377334894384,
+            8.741009867015967e-05,
+            0.0007286815550017811,
+            0.00019660620632828166,
+        ],
+        'total_input_samples': 3,
+        'valid_input_samples': 3,
+        'model_task': 'BINARY_CLASSIFICATION',
+        'model_input_type': 'TABULAR',
+        'env_id': 'a6385884-4970-4eeb-b8cc-36ce9746cd5e',
+        'env_name': DATASET_NAME,
+        'created_at': '2023-11-22 16:50:57.705784',
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+FEATURE_IMPACT_RESPONSE_200 = {
+    'data': {
+        'num_inputs': 3,
+        'num_refs': 10000,
+        'ci_level': 0.5,
+        'mean_prediction': 0.08983428988388353,
+        'mean_prediction_ci': 0.006687591607821416,
+        'feature_names': [
+            'CreditScore',
+            'Geography',
+            'Gender',
+            'Age',
+            'Tenure',
+            'Balance',
+            'NumOfProducts',
+            'HasCrCard',
+            'IsActiveMember',
+            'EstimatedSalary',
+        ],
+        'mean_abs_prediction_change_impact': [
+            0.00712590582097325,
+            0.04386081743167407,
+            0.01618882705936992,
+            0.09012170216686392,
+            0.003203886657934025,
+            0.034259594581994,
+            0.03950092221577272,
+            0.0008308543417424118,
+            0.09500517056814688,
+            0.015902298170793795,
+        ],
+        'random_sample_ci': [
+            9.460265530465104e-05,
+            0.0004967738708175746,
+            0.0001390532937646932,
+            0.001011710786484173,
+            3.2151564756512305e-05,
+            0.0003379320087881252,
+            0.0007824117094712736,
+            6.632377333582839e-06,
+            0.001290117285928614,
+            0.0001246149963496527,
+        ],
+        'fixed_sample_ci': [
+            9.387290704313362e-05,
+            0.00047072988314235967,
+            0.00011810560126585543,
+            0.0009869435854980898,
+            2.6313946023718767e-05,
+            0.0003171540654523467,
+            0.0007768114901140053,
+            5.596843413373775e-06,
+            0.001029967633853864,
+            0.00011703590760866858,
+        ],
+        'model_task': 'BINARY_CLASSIFICATION',
+        'model_input_type': 'TABULAR',
+        'env_id': 'a6385884-4970-4eeb-b8cc-36ce9746cd5e',
+        'env_name': DATASET_NAME,
+        'created_at': '2023-11-22 16:50:57.705784',
+    },
+    'api_version': '3.0',
+    'kind': 'NORMAL',
+}
+
+
+@responses.activate
+def test_explain() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/explain',
+        json=EXPLAIN_RESPONSE_200,
+    )
+    explain_result = model.explain(
+        input_data_source=RowDataSource(
+            row={
+                'CreditScore': 619,
+                'Geography': 'France',
+                'Gender': 'Female',
+                'Age': 42,
+                'Tenure': 2,
+                'Balance': 0.0,
+                'NumOfProducts': 1,
+                'HasCrCard': 'Yes',
+                'IsActiveMember': 'Yes',
+                'EstimatedSalary': 101348.88,
+            },
+        ),
+        ref_data_source=DatasetDataSource(
+            env_type='PRODUCTION',
+        ),
+    )
+    assert explain_result == namedtuple('Explain', EXPLAIN_RESPONSE_200['data'])(
+        **EXPLAIN_RESPONSE_200['data']
+    )
+
+
+@responses.activate
+def test_fairness() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/analytics/fairness',
+        json=FAIRNESS_RESPONSE_200,
+    )
+    fairness = model.get_fairness(
+        data_source=DatasetDataSource(
+            env_type='PRODUCTION',
+        ),
+        protected_features=['Gender'],
+        positive_outcome='Churned',
+    )
+    assert fairness == namedtuple('Fairness', FAIRNESS_RESPONSE_200['data'])(
+        **FAIRNESS_RESPONSE_200['data']
+    )
+
+
+@responses.activate
+def test_get_slice() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/slice-query/fetch',
+        json=FETCH_SLICE_RESPONSE_200,
+    )
+
+    slice_df = model.get_slice(
+        query="SELECT * FROM test_bank_churn.bank_churn WHERE geography='France' order by balance desc LIMIT 3",
+        columns=['Age'],
+        max_rows=2,
+    )
+    expected_df = pd.DataFrame({'Age': [57, 37]})
+    pd.testing.assert_frame_equal(expected_df, slice_df)
+
+
+@responses.activate
+def test_slice_download() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    parquet_path = os.path.join(OUTPUT_DIR, 'test_slice_download.parquet')
+    parquet_output = os.path.join(BASE_TEST_DIR, 'slice_test_dir')
+    with open(parquet_path, 'rb') as parquet_file:
+        data = io.BufferedReader(parquet_file)
+        responses.post(
+            url=f'{URL}/v3/slice-query/download',
+            body=data,
+        )
+        expected_df = pd.DataFrame({'Age': [38, 57, 42]})
+        model.download_slice(
+            output_dir=parquet_output,
+            query='SELECT * FROM test_bank_churn.bank_churn WHERE age>=20 order by balance desc',
+            columns=['Age'],
+            max_rows=3,
+        )
+        slice_df = pd.read_parquet(parquet_path)
+        pd.testing.assert_frame_equal(expected_df, slice_df)
+    shutil.rmtree(str(parquet_output))
+
+
+@responses.activate
+def test_mutual_info() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/analytics/mutual-info',
+        json=MUTUAL_INFO_RESPONSE_200,
+    )
+    mutual_info = model.get_mutual_info(
+        query=f'select * from {DATASET_NAME}.{MODEL_NAME}',
+        column_name='Geography',
+    )
+    assert mutual_info == MUTUAL_INFO_RESPONSE_200['data']
+
+
+@responses.activate
+def test_predict() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/predict',
+        json=PREDICT_RESPONSE_200,
+    )
+    data = {
+        'row_id': 1109,
+        'fixed acidity': 10.8,
+        'volatile acidity': 0.47,
+        'citric acid': 0.43,
+        'residual sugar': 2.1,
+        'chlorides': 0.171,
+        'free sulfur dioxide': 27.0,
+        'total sulfur dioxide': 66.0,
+        'density': 0.9982,
+        'pH': 3.17,
+        'sulphates': 0.76,
+        'alcohol': 10.8,
+    }
+    df = pd.DataFrame(data, index=data.keys())
+    predictions = model.predict(df=df)
+
+    expected_result = pd.DataFrame(PREDICT_RESPONSE_200['data']['predictions'])
+    pd.testing.assert_frame_equal(predictions, expected_result)
+
+
+@responses.activate
+def test_get_feature_importance() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/analytics/feature-importance/precomputed',
+        json=FEATURE_IMPORTANCE_RESPONSE_200,
+    )
+    assert model.get_precomputed_feature_importance() == namedtuple(
+        'FeatureImportance', FEATURE_IMPORTANCE_RESPONSE_200['data']
+    )(**FEATURE_IMPORTANCE_RESPONSE_200['data'])
+
+
+@responses.activate
+def test_get_feature_impact() -> None:
+    responses.get(
+        url=f'{URL}/v3/models/{MODEL_ID}',
+        json=MODEL_API_RESPONSE_200,
+    )
+    model = Model.get(id_=MODEL_ID)
+
+    responses.post(
+        url=f'{URL}/v3/analytics/feature-impact/precomputed',
+        json=FEATURE_IMPACT_RESPONSE_200,
+    )
+    assert model.get_precomputed_feature_impact() == namedtuple(
+        'FeatureImpact', FEATURE_IMPACT_RESPONSE_200['data']
+    )(**FEATURE_IMPACT_RESPONSE_200['data'])

fiddler3/tests/constants.py

@@ -1,3 +1,6 @@
+import os
+from pathlib import Path
+
 URL = 'https://dev.fiddler.ai'
 TOKEN = 'footoken'
 ORG_ID = '5531bfd9-2ca2-4a7b-bb5a-136c8da09ca0'
@@ -12,3 +15,10 @@ USER_NAME = 'foo'
 USER_EMAIL = 'foo@fiddler.ai'
 JOB_ID = '9df70575-7ced-42fd-bb50-a0e4351d2c9d'
 JOB_NAME = f'Deleting model {MODEL_NAME}'
+DATASET_ID = 'ba6ec4e4-7188-44c5-ba84-c2cb22b4bb00'
+DATASET_NAME = 'dataset3'
+BASELINE_ID = 'af05646f-0cef-4638-84c9-0d195df2575d'
+BASELINE_NAME = 'test_baseline'
+MODEL_DEPLOYMENT_ID = 'fc07562a-1a27-4f66-9154-732d3b14b98a'
+BASE_TEST_DIR = f'{Path(__file__).resolve().parent}'
+OUTPUT_DIR = os.path.join(BASE_TEST_DIR, 'artifact_test_dir')