fix: address bigquery/bqml test failures (#2920)

* fix: bigquery/bqml testing The BQML tests use a non-unique dataset ID for multiple examples and testing currently triggers a lot of concurrent creation/deletions of said dataset. Switch to a dataset that leverages uuid to avoid invocations stomping on one another, which also necessitates parameterizing much of the SQL. There's also an issue with the pandas import currently, possibly due to recent changes in panda. This change also pins pandas to 0.22 and doesn't rely on the dependency being expressed as an extra through google-cloud-bigquery. * whitespace lint * update dependencies in requirements.txt
GoogleCloudPlatform · Feb 15, 2020 · 58752d5 · 58752d5
1 parent 5ca361b
commit 58752d5
Show file tree

Hide file tree

Showing 5 changed files with 50 additions and 40 deletions.
diff --git a/bigquery/bqml/data_scientist_tutorial_test.py b/bigquery/bqml/data_scientist_tutorial_test.py
@@ -16,29 +16,34 @@
 from google.cloud import bigquery
 # [END bqml_data_scientist_tutorial_import_and_client]
 import pytest
+import uuid
 
 # [START bqml_data_scientist_tutorial_import_and_client]
 client = bigquery.Client()
+# We use a unique dataset ID for this example to avoid collisions with
+# other invocations of this tutorial.  In practice, you could leverage
+# a persistent dataset and not create/destroy it with each invocation.
+dataset_id = "bqml_tutorial_{}".format(str(uuid.uuid4().hex))
 # [END bqml_data_scientist_tutorial_import_and_client]
 
 
 @pytest.fixture
 def delete_dataset():
     yield
     client.delete_dataset(
-        client.dataset('bqml_tutorial'), delete_contents=True)
+        client.dataset(dataset_id), delete_contents=True)
 
 
 def test_data_scientist_tutorial(delete_dataset):
     # [START bqml_data_scientist_tutorial_create_dataset]
-    dataset = bigquery.Dataset(client.dataset('bqml_tutorial'))
+    dataset = bigquery.Dataset(client.dataset(dataset_id))
     dataset.location = 'US'
     client.create_dataset(dataset)
     # [END bqml_data_scientist_tutorial_create_dataset]
 
     # [START bqml_data_scientist_tutorial_create_model]
     sql = """
-        CREATE OR REPLACE MODEL `bqml_tutorial.sample_model`
+        CREATE OR REPLACE MODEL `{}.sample_model`
         OPTIONS(model_type='logistic_reg') AS
         SELECT
             IF(totals.transactions IS NULL, 0, 1) AS label,
@@ -50,7 +55,7 @@ def test_data_scientist_tutorial(delete_dataset):
             `bigquery-public-data.google_analytics_sample.ga_sessions_*`
         WHERE
             _TABLE_SUFFIX BETWEEN '20160801' AND '20170630'
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_data_scientist_tutorial_create_model]
@@ -60,8 +65,8 @@ def test_data_scientist_tutorial(delete_dataset):
         SELECT
         *
         FROM
-        ML.TRAINING_INFO(MODEL `bqml_tutorial.sample_model`)
-    """
+        ML.TRAINING_INFO(MODEL `{}.sample_model`)
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_data_scientist_tutorial_get_training_statistics]
@@ -70,7 +75,7 @@ def test_data_scientist_tutorial(delete_dataset):
     sql = """
         SELECT
             *
-        FROM ML.EVALUATE(MODEL `bqml_tutorial.sample_model`, (
+        FROM ML.EVALUATE(MODEL `{}.sample_model`, (
             SELECT
                 IF(totals.transactions IS NULL, 0, 1) AS label,
                 IFNULL(device.operatingSystem, "") AS os,
@@ -81,7 +86,7 @@ def test_data_scientist_tutorial(delete_dataset):
                 `bigquery-public-data.google_analytics_sample.ga_sessions_*`
             WHERE
                 _TABLE_SUFFIX BETWEEN '20170701' AND '20170801'))
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_data_scientist_tutorial_evaluate_model]
@@ -91,7 +96,7 @@ def test_data_scientist_tutorial(delete_dataset):
         SELECT
             country,
             SUM(predicted_label) as total_predicted_purchases
-        FROM ML.PREDICT(MODEL `bqml_tutorial.sample_model`, (
+        FROM ML.PREDICT(MODEL `{}.sample_model`, (
             SELECT
                 IFNULL(device.operatingSystem, "") AS os,
                 device.isMobile AS is_mobile,
@@ -104,7 +109,7 @@ def test_data_scientist_tutorial(delete_dataset):
             GROUP BY country
             ORDER BY total_predicted_purchases DESC
             LIMIT 10
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_data_scientist_tutorial_predict_transactions]
@@ -114,7 +119,7 @@ def test_data_scientist_tutorial(delete_dataset):
         SELECT
             fullVisitorId,
             SUM(predicted_label) as total_predicted_purchases
-        FROM ML.PREDICT(MODEL `bqml_tutorial.sample_model`, (
+        FROM ML.PREDICT(MODEL `{}.sample_model`, (
             SELECT
                 IFNULL(device.operatingSystem, "") AS os,
                 device.isMobile AS is_mobile,
@@ -128,7 +133,7 @@ def test_data_scientist_tutorial(delete_dataset):
             GROUP BY fullVisitorId
             ORDER BY total_predicted_purchases DESC
             LIMIT 10
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_data_scientist_tutorial_predict_purchases]
diff --git a/bigquery/bqml/ncaa_tutorial_test.py b/bigquery/bqml/ncaa_tutorial_test.py
@@ -14,6 +14,7 @@
 
 import io
 import os
+import uuid
 
 # [START bqml_ncaa_tutorial_import_and_client]
 from google.cloud import bigquery
@@ -22,43 +23,41 @@
 
 # [START bqml_ncaa_tutorial_import_and_client]
 client = bigquery.Client()
+# We use a unique dataset ID for this example to avoid collisions with
+# other invocations of this tutorial.  In practice, you could leverage
+# a persistent dataset and not create/destroy it with each invocation.
+dataset_id = "bqml_tutorial_{}".format(str(uuid.uuid4().hex))
 # [END bqml_ncaa_tutorial_import_and_client]
 
 
 @pytest.fixture
 def delete_dataset():
     yield
     client.delete_dataset(
-        client.dataset('bqml_tutorial'), delete_contents=True)
+        client.dataset(dataset_id), delete_contents=True)
 
 
 def test_ncaa_tutorial(delete_dataset):
     # [START bqml_ncaa_tutorial_create_dataset]
-    dataset = bigquery.Dataset(client.dataset('bqml_tutorial'))
+    dataset = bigquery.Dataset(client.dataset(dataset_id))
     dataset.location = 'US'
     client.create_dataset(dataset)
     # [END bqml_ncaa_tutorial_create_dataset]
 
     # Create the tables used by the tutorial
     # Note: the queries are saved to a file. This should be updated to use the
     # saved queries once the library supports running saved queries.
-    query_filepath_to_table_name = {
-        'feature_input_query.sql': 'cume_games',
-        'training_data_query.sql': 'wide_games'
-    }
+    query_files = ['feature_input_query.sql', 'training_data_query.sql']
     resources_directory = os.path.join(os.path.dirname(__file__), 'resources')
-    for query_filepath, table_name in query_filepath_to_table_name.items():
-        table_ref = dataset.table(table_name)
-        job_config = bigquery.QueryJobConfig()
-        job_config.destination = table_ref
+    for fname in query_files:
         query_filepath = os.path.join(
-            resources_directory, query_filepath)
-        sql = io.open(query_filepath, 'r', encoding='utf-8').read()
-        client.query(sql, job_config=job_config).result()
+            resources_directory, fname)
+        sql = io.open(query_filepath, 'r', encoding='utf-8').read().format(dataset_id)
+        client.query(sql).result()
 
     # [START bqml_ncaa_tutorial_create_model]
     sql = """
-        CREATE OR REPLACE MODEL `bqml_tutorial.ncaa_model`
+        CREATE OR REPLACE MODEL `{0}.ncaa_model`
         OPTIONS (
             model_type='linear_reg',
             max_iteration=50 ) AS
@@ -69,11 +68,11 @@ def test_ncaa_tutorial(delete_dataset):
                 total_three_points_att),
             total_three_points_att as label
         FROM
-            `bqml_tutorial.wide_games`
+            `{0}.wide_games`
         WHERE
             # remove the game to predict
             game_id != 'f1063e80-23c7-486b-9a5e-faa52beb2d83'
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_ncaa_tutorial_create_model]
@@ -83,8 +82,8 @@ def test_ncaa_tutorial(delete_dataset):
         SELECT
             *
         FROM
-            ML.TRAINING_INFO(MODEL `bqml_tutorial.ncaa_model`)
-    """
+            ML.TRAINING_INFO(MODEL `{}.ncaa_model`)
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_ncaa_tutorial_get_training_statistics]
@@ -96,13 +95,13 @@ def test_ncaa_tutorial(delete_dataset):
                 *,
                 total_three_points_att AS label
             FROM
-                `bqml_tutorial.wide_games` )
+                `{0}.wide_games` )
         SELECT
             *
         FROM
-            ML.EVALUATE(MODEL `bqml_tutorial.ncaa_model`,
+            ML.EVALUATE(MODEL `{0}.ncaa_model`,
                 TABLE eval_table)
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_ncaa_tutorial_evaluate_model]
@@ -113,7 +112,7 @@ def test_ncaa_tutorial(delete_dataset):
             SELECT
                 *
             FROM
-                `bqml_tutorial.wide_games`
+                `{0}.wide_games`
             WHERE
                 game_id='f1063e80-23c7-486b-9a5e-faa52beb2d83' )
         SELECT
@@ -125,7 +124,7 @@ def test_ncaa_tutorial(delete_dataset):
                 game_id,
                 predicted_label AS predicted_total_three_points_att
             FROM
-                ML.PREDICT(MODEL `bqml_tutorial.ncaa_model`,
+                ML.PREDICT(MODEL `{0}.ncaa_model`,
                 table game_to_predict) ) AS predict
         JOIN (
             SELECT
@@ -135,7 +134,7 @@ def test_ncaa_tutorial(delete_dataset):
                 game_to_predict) AS truth
         ON
             predict.game_id = truth.game_id
-    """
+    """.format(dataset_id)
     df = client.query(sql).to_dataframe()
     print(df)
     # [END bqml_ncaa_tutorial_predict_outcomes]
diff --git a/bigquery/bqml/requirements.txt b/bigquery/bqml/requirements.txt
@@ -1,3 +1,4 @@
-google-cloud-bigquery[pandas]==1.20.0
+pandas==0.24.2
+google-cloud-bigquery==1.23.1
 flaky==3.6.1
 mock==3.0.5
diff --git a/bigquery/bqml/resources/feature_input_query.sql b/bigquery/bqml/resources/feature_input_query.sql
@@ -1,4 +1,8 @@
-#standardSQL
+# This query creates a sample table using
+# the ncaa_basketball public dataset.  It
+# uses a format string token for setting
+# the destination dataset.
+CREATE OR REPLACE TABLE `{0}.cume_games` AS
 SELECT
   game_id,
   season,

diff --git a/bigquery/bqml/resources/training_data_query.sql b/bigquery/bqml/resources/training_data_query.sql
@@ -1,4 +1,5 @@
 #standardSQL
+CREATE OR REPLACE TABLE `{0}.wide_games` AS
 SELECT
   team.game_id AS game_id,
   team.season AS season,
@@ -768,9 +769,9 @@ SELECT
   opponent.opp_possessions_std_last_5 AS	opponent_opp_possessions_std_last_5,
   opponent.opp_possessions_std_last_10 AS	opponent_opp_possessions_std_last_10
 FROM
-  `bqml_tutorial.cume_games` AS team
+  `{0}.cume_games` AS team
 JOIN
-  `bqml_tutorial.cume_games` AS opponent
+  `{0}.cume_games` AS opponent
 ON
   team.game_id = opponent.game_id AND team.team_id != opponent.team_id
 WHERE