Skip to content

Commit 6a3b0cc

Browse files
DevStephanietswast
DevStephanie
and
tswast
authored
docs: add predict sample to samples/snippets/bqml_getting_started_test.py (#388)
* docs: Add a sample to demonstrate the evaluation results * Adding comments explaining logistic regression results * editing read_gbd explanation * docs: add predict sample to samples/snippets/bqml_getting_started_test.py * correcting variable names * Correcting python variables * feat: add predict by visit to samples/snippets/bqml_getting_started_test.py * file * file * file --------- Co-authored-by: Tim Sweña (Swast, formerly) <swast@google.com>
1 parent 11a3743 commit 6a3b0cc

File tree

1 file changed

+171
-42
lines changed

1 file changed

+171
-42
lines changed

‎samples/snippets/bqml_getting_started_test.py‎

Lines changed: 171 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,7 @@
1414

1515

1616
def test_bqml_getting_started(random_model_id):
17-
your_model_id = random_model_id
17+
your_model_id = random_model_id # for example: bqml_tutorial.sample_model
1818

1919
# [START bigquery_dataframes_bqml_getting_started_tutorial]
2020
from bigframes.ml.linear_model import LogisticRegression
@@ -26,17 +26,12 @@ def test_bqml_getting_started(random_model_id):
2626
# https://github.com/googleapis/python-bigquery-dataframes/issues/169
2727
# for updates to `read_gbq` to support wildcard tables.
2828

29-
df = bpd.read_gbq(
30-
"""
31-
-- Since the order of rows isn't useful for the model training,
32-
-- generate a random ID to use as the index for the DataFrame.
33-
SELECT GENERATE_UUID() AS rowindex, *
34-
FROM
35-
`bigquery-public-data.google_analytics_sample.ga_sessions_*`
36-
WHERE
37-
_TABLE_SUFFIX BETWEEN '20160801' AND '20170630'
38-
""",
39-
index_col="rowindex",
29+
df = bpd.read_gbq_table(
30+
"bigquery-public-data.google_analytics_sample.ga_sessions_*",
31+
filters=[
32+
("_table_suffix", ">=", "20160801"),
33+
("_table_suffix", "<=", "20170630"),
34+
],
4035
)
4136

4237
# Extract the total number of transactions within
@@ -53,14 +48,14 @@ def test_bqml_getting_started(random_model_id):
5348
# ecommerce transactions within the Google Analytics session.
5449
# If the number of transactions is NULL, the value in the label
5550
# column is set to 0. Otherwise, it is set to 1.
56-
label = transactions.notnull().map({True: 1, False: 0})
51+
label = transactions.notnull().map({True: 1, False: 0}).rename("label")
5752

5853
# Extract the operating system of the visitor's device.
59-
operatingSystem = df["device"].struct.field("operatingSystem")
60-
operatingSystem = operatingSystem.fillna("")
54+
operating_system = df["device"].struct.field("operatingSystem")
55+
operating_system = operating_system.fillna("")
6156

6257
# Extract whether the visitor's device is a mobile device.
63-
isMobile = df["device"].struct.field("isMobile")
58+
is_mobile = df["device"].struct.field("isMobile")
6459

6560
# Extract the country from which the sessions originated, based on the IP address.
6661
country = df["geoNetwork"].struct.field("country").fillna("")
@@ -72,8 +67,8 @@ def test_bqml_getting_started(random_model_id):
7267
# to use as training data.
7368
features = bpd.DataFrame(
7469
{
75-
"os": operatingSystem,
76-
"is_mobile": isMobile,
70+
"os": operating_system,
71+
"is_mobile": is_mobile,
7772
"country": country,
7873
"pageviews": pageviews,
7974
}
@@ -95,39 +90,36 @@ def test_bqml_getting_started(random_model_id):
9590
# [START bigquery_dataframes_bqml_getting_started_tutorial_evaluate]
9691
import bigframes.pandas as bpd
9792

98-
# Select model you'll use for training. `read_gbq_model` loads model data from a
93+
# Select model you'll use for evaluating. `read_gbq_model` loads model data from a
9994
# BigQuery, but you could also use the `model` object from the previous steps.
10095
model = bpd.read_gbq_model(
10196
your_model_id, # For example: "bqml_tutorial.sample_model",
10297
)
10398

104-
# The WHERE clause — _TABLE_SUFFIX BETWEEN '20170701' AND '20170801' —
105-
# limits the number of tables scanned by the query. The date range scanned is
106-
# July 1, 2017 to August 1, 2017. This is the data you're using to evaluate the predictive performance
107-
# of the model. It was collected in the month immediately following the time
108-
# period spanned by the training data.
109-
110-
df = bpd.read_gbq(
111-
"""
112-
SELECT GENERATE_UUID() AS rowindex, *
113-
FROM
114-
`bigquery-public-data.google_analytics_sample.ga_sessions_*`
115-
WHERE
116-
_TABLE_SUFFIX BETWEEN '20170701' AND '20170801'
117-
""",
118-
index_col="rowindex",
99+
# The filters parameter limits the number of tables scanned by the query.
100+
# The date range scanned is July 1, 2017 to August 1, 2017. This is the
101+
# data you're using to evaluate the predictive performance of the model.
102+
# It was collected in the month immediately following the time period
103+
# spanned by the training data.
104+
df = bpd.read_gbq_table(
105+
"bigquery-public-data.google_analytics_sample.ga_sessions_*",
106+
filters=[
107+
("_table_suffix", ">=", "20170701"),
108+
("_table_suffix", "<=", "20170801"),
109+
],
119110
)
111+
120112
transactions = df["totals"].struct.field("transactions")
121-
label = transactions.notnull().map({True: 1, False: 0})
122-
operatingSystem = df["device"].struct.field("operatingSystem")
123-
operatingSystem = operatingSystem.fillna("")
124-
isMobile = df["device"].struct.field("isMobile")
113+
label = transactions.notnull().map({True: 1, False: 0}).rename("label")
114+
operating_system = df["device"].struct.field("operatingSystem")
115+
operating_system = operating_system.fillna("")
116+
is_mobile = df["device"].struct.field("isMobile")
125117
country = df["geoNetwork"].struct.field("country").fillna("")
126118
pageviews = df["totals"].struct.field("pageviews").fillna(0)
127119
features = bpd.DataFrame(
128120
{
129-
"os": operatingSystem,
130-
"is_mobile": isMobile,
121+
"os": operating_system,
122+
"is_mobile": is_mobile,
131123
"country": country,
132124
"pageviews": pageviews,
133125
}
@@ -163,6 +155,143 @@ def test_bqml_getting_started(random_model_id):
163155
# [1 rows x 6 columns]
164156
# [END bigquery_dataframes_bqml_getting_started_tutorial_evaluate]
165157

166-
# [START bigquery_dataframes_bqml_getting_started_tutorial_predict]
158+
# [START bigquery_dataframes_bqml_getting_started_tutorial_predict_by_country]
159+
import bigframes.pandas as bpd
160+
161+
# Select model you'll use for predicting.
162+
# `read_gbq_model` loads model data from
163+
# BigQuery, but you could also use the `model`
164+
# object from the previous steps.
165+
model = bpd.read_gbq_model(
166+
your_model_id, # For example: "bqml_tutorial.sample_model",
167+
)
168+
169+
# The filters parameter limits the number of tables scanned by the query.
170+
# The date range scanned is July 1, 2017 to August 1, 2017. This is the
171+
# data you're using to make the prediction.
172+
# It was collected in the month immediately following the time period
173+
# spanned by the training data.
174+
df = bpd.read_gbq_table(
175+
"bigquery-public-data.google_analytics_sample.ga_sessions_*",
176+
filters=[
177+
("_table_suffix", ">=", "20170701"),
178+
("_table_suffix", "<=", "20170801"),
179+
],
180+
)
181+
182+
operating_system = df["device"].struct.field("operatingSystem")
183+
operating_system = operating_system.fillna("")
184+
is_mobile = df["device"].struct.field("isMobile")
185+
country = df["geoNetwork"].struct.field("country").fillna("")
186+
pageviews = df["totals"].struct.field("pageviews").fillna(0)
187+
features = bpd.DataFrame(
188+
{
189+
"os": operating_system,
190+
"is_mobile": is_mobile,
191+
"country": country,
192+
"pageviews": pageviews,
193+
}
194+
)
195+
# Use Logistic Regression predict method to predict results
196+
# using your model.
197+
# Find more information here in
198+
# [BigFrames](https://cloud.google.com/python/docs/reference/bigframes/latest/bigframes.ml.linear_model.LogisticRegression#bigframes_ml_linear_model_LogisticRegression_predict)
199+
200+
predictions = model.predict(features)
201+
202+
# Call groupby method to group predicted_label by country.
203+
# Call sum method to get the total_predicted_label by country.
204+
total_predicted_purchases = predictions.groupby(["country"])[
205+
["predicted_label"]
206+
].sum()
207+
208+
# Call the sort_values method with the parameter
209+
# ascending = False to get the highest values.
210+
# Call head method to limit to the 10 highest values.
211+
total_predicted_purchases.sort_values(ascending=False).head(10)
212+
213+
# country
214+
# United States 220
215+
# Taiwan 8
216+
# Canada 7
217+
# India 2
218+
# Japan 2
219+
# Turkey 2
220+
# Australia 1
221+
# Brazil 1
222+
# Germany 1
223+
# Guyana 1
224+
# Name: predicted_label, dtype: Int64
225+
226+
# [END bigquery_dataframes_bqml_getting_started_tutorial_predict_by_country]
227+
228+
# [START bigquery_dataframes_bqml_getting_started_tutorial_predict_by_visitor]
229+
230+
import bigframes.pandas as bpd
231+
232+
# Select model you'll use for predicting.
233+
# `read_gbq_model` loads model data from
234+
# BigQuery, but you could also use the `model`
235+
# object from the previous steps.
236+
model = bpd.read_gbq_model(
237+
your_model_id, # For example: "bqml_tutorial.sample_model",
238+
)
239+
240+
# The filters parameter limits the number of tables scanned by the query.
241+
# The date range scanned is July 1, 2017 to August 1, 2017. This is the
242+
# data you're using to make the prediction.
243+
# It was collected in the month immediately following the time period
244+
# spanned by the training data.
245+
df = bpd.read_gbq_table(
246+
"bigquery-public-data.google_analytics_sample.ga_sessions_*",
247+
filters=[
248+
("_table_suffix", ">=", "20170701"),
249+
("_table_suffix", "<=", "20170801"),
250+
],
251+
)
252+
253+
operating_system = df["device"].struct.field("operatingSystem")
254+
operating_system = operating_system.fillna("")
255+
is_mobile = df["device"].struct.field("isMobile")
256+
country = df["geoNetwork"].struct.field("country").fillna("")
257+
pageviews = df["totals"].struct.field("pageviews").fillna(0)
258+
full_visitor_id = df["fullVisitorId"]
259+
260+
features = bpd.DataFrame(
261+
{
262+
"os": operating_system,
263+
"is_mobile": is_mobile,
264+
"country": country,
265+
"pageviews": pageviews,
266+
"fullVisitorId": full_visitor_id,
267+
}
268+
)
269+
270+
predictions = model.predict(features)
271+
272+
# Call groupby method to group predicted_label by visitor.
273+
# Call sum method to get the total_predicted_label by visitor.
274+
total_predicted_purchases = predictions.groupby(["fullVisitorId"])[
275+
["predicted_label"]
276+
].sum()
277+
278+
# Call the sort_values method with the parameter
279+
# ascending = False to get the highest values.
280+
# Call head method to limit to the 10 highest values.
281+
total_predicted_purchases.sort_values(ascending=False).head(10)
282+
283+
# fullVisitorId
284+
# 9417857471295131045 4
285+
# 0376394056092189113 2
286+
# 0456807427403774085 2
287+
# 057693500927581077 2
288+
# 112288330928895942 2
289+
# 1280993661204347450 2
290+
# 2105122376016897629 2
291+
# 2158257269735455737 2
292+
# 2969418676126258798 2
293+
# 489038402765684003 2
294+
# Name: predicted_label, dtype: Int64
295+
167296

168-
# [END bigquery_dataframes_bqml_getting_started_tutorial_predict]
297+
# [END bigquery_dataframes_bqml_getting_started_tutorial_predict_by_visitor]

0 commit comments

Comments
 (0)