-
Notifications
You must be signed in to change notification settings - Fork 27
/
etl_job.py
299 lines (247 loc) · 9.06 KB
/
etl_job.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
"""Used as part of AWS Glue Job to transform credits data."""
# see https://github.com/awslabs/aws-glue-libs/tree/master/awsglue for details
from awsglue.context import GlueContext # pylint: disable=import-error
from awsglue.job import Job # pylint: disable=import-error
from awsglue.utils import getResolvedOptions # pylint: disable=import-error
import boto3
from pyspark import SparkConf, SparkContext
import sys
def setup_contexts():
spark_conf = (
SparkConf()
.set(
"spark.hadoop.hive.metastore.client.factory.class",
"com.amazonaws.glue.catalog.metastore"
+ ".AWSGlueDataCatalogHiveClientFactory",
)
.set("spark.sql.catalogImplementation", "hive")
)
sc = SparkContext(conf=spark_conf)
glueContext = GlueContext(sc)
spark = glueContext.spark_session
return glueContext, spark
def create_job(glueContext, args):
job = Job(glueContext)
job.init(args["JOB_NAME"], args)
return job
def get_database_location(database_name, region):
glue = boto3.client("glue", region_name=region)
database = glue.get_database(Name=database_name)
assert "Database" in database
assert (
"LocationUri" in database["Database"]
), "Must set LocationUri on database."
location_uri = database["Database"]["LocationUri"]
s3_bucket, s3_prefix = parse_database_location(location_uri)
return s3_bucket, s3_prefix
def parse_database_location(location_uri):
assert (
location_uri[:5] == "s3://"
), "database location expected to be an s3 uri."
s3_bucket = location_uri[5:].split("/")[0]
s3_prefix = "/".join(location_uri[5:].split("/")[1:]).strip("/")
return s3_bucket, s3_prefix
def load_credits():
credits = glueContext.create_dynamic_frame.from_catalog(
database=GLUE_DATABASE, table_name="credits"
)
return credits
def transform_credits(credits):
credits = credits.toDF() # convert from dynamic frame to data frame
return credits
def load_people():
people = glueContext.create_dynamic_frame.from_catalog(
database=GLUE_DATABASE, table_name="people"
)
return people
def relationalize_people(people):
# flatten and unnest attributes
dfc = people.relationalize("people", GLUE_TEMP)
# `dfc` is a 'dynamic frame collection' with two frames
# `people` contains the flattened attributes
people = dfc.select("people").toDF()
people = people.toDF(
*[c.replace(".", "__") for c in people.columns]
) # avoid issues with '.'
dependents = dfc.select("people_dependents").toDF()
return people, dependents
def fill_finance_accounts(people):
people = people.na.fill(
{"finance__accounts__checking__balance": "no_account"}
)
people = people.na.fill(
{"finance__accounts__savings__balance": "no_account"}
)
return people
def join_dependents(people, dependents):
# create a new feature called `num_dependents`
people.createOrReplaceTempView("people")
dependents.createOrReplaceTempView("dependents")
num_dependents = spark.sql(
"""
SELECT id, count(id) as num_dependents
FROM dependents
GROUP BY id
"""
)
# add this feature to `people` table
num_dependents.createOrReplaceTempView("num_dependents")
people = spark.sql(
"""
SELECT *
FROM people
LEFT JOIN num_dependents ON people.dependents=num_dependents.id
"""
)
# clean up columns
people = people.drop(*["dependents", "id"])
people = people.withColumnRenamed(
"num_dependents", "personal__num_dependents"
)
return people
def transform_people(people):
people, dependents = relationalize_people(people)
people = fill_finance_accounts(people)
people = join_dependents(people, dependents)
# `people_dependents` contains the previously nested array of dependents
return people
def load_contacts():
contacts = glueContext.create_dynamic_frame.from_catalog(
database=GLUE_DATABASE, table_name="contacts"
)
return contacts
def transform_contacts(contacts):
contacts.toDF().createOrReplaceTempView("contacts")
contacts = spark.sql(
"""
SELECT person_id, count(contact_id) as num_telephones
FROM contacts
WHERE type = 'telephone'
GROUP BY person_id
"""
)
return contacts
def join_data(credits, people, contacts):
# rename before join to avoid column name clash
credits = credits.toDF(*["credit__" + n for n in credits.columns])
contacts = contacts.toDF(*["contact__" + n for n in contacts.columns])
# add temp views so can access from spark sql
credits.createOrReplaceTempView("credits")
people.createOrReplaceTempView("people")
contacts.createOrReplaceTempView("contacts")
# join all tables on `person_id`
data = spark.sql(
"""
SELECT *
FROM credits
JOIN people ON credits.credit__person_id = people.person_id
LEFT JOIN contacts ON credits.credit__person_id =
contacts.contact__person_id
"""
)
return data
def transform_data(data):
# clean up columns
data = data.selectExpr(
"*", "isnotnull(contact__num_telephones) as contact__has_telephone"
)
data = data.drop("contact__num_telephones")
data = data.drop("credit__credit_id")
data = data.drop(*["credit__person_id", "person_id", "contact__person_id"])
data = data.drop(*["employment__title", "personal__name"])
data = data.select(["`{}`".format(c) for c in sorted(data.columns)])
return data
def split_train_test(data, train_split=0.8):
data = data.cache()
data_train, data_test = data.randomSplit([train_split, 1 - train_split])
return data_train, data_test
def slice_protected(data_train, data_test):
protected = [
"personal__age",
"personal__gender",
"personal__relationship_status",
]
protected_train = data_train.select(protected)
data_train = data_train.drop(*protected)
protected_test = data_test.select(protected)
data_test = data_test.drop(*protected)
return data_train, protected_train, data_test, protected_test
def slice_label(data_train, data_test):
label = "credit__default"
label_train = data_train.select(label)
data_train = data_train.drop(label)
label_test = data_test.select(label)
data_test = data_test.drop(label)
return data_train, label_train, data_test, label_test
def delete_table_data(table_name):
s3 = boto3.resource("s3")
bucket = s3.Bucket(S3_BUCKET) # pylint: disable=no-member
bucket.objects.filter(
Prefix="{}/{}/".format(S3_PREFIX, table_name)
).delete()
def delete_table(table_name):
spark.sql("DROP TABLE IF EXISTS `{}`".format(table_name))
def create_table(df, table_name):
df.registerTempTable("df_tmp")
spark.sql(
"""
CREATE TABLE IF NOT EXISTS {}
USING HIVE
OPTIONS(
fileFormat 'textfile',
serde 'org.openx.data.jsonserde.JsonSerDe')
AS SELECT * FROM df_tmp
""".format(
table_name
)
)
def save(df, table_name, overwrite=False):
if overwrite:
delete_table(table_name)
delete_table_data(table_name)
create_table(df, table_name)
def main():
# set default database to avoid having to reference each time
spark.sql("USE `{}`".format(GLUE_DATABASE))
# load and transform each of the three tables
credits = load_credits()
credits = transform_credits(credits)
people = load_people()
people = transform_people(people)
contacts = load_contacts()
contacts = transform_contacts(contacts)
# join all three tables into one
data = join_data(credits, people, contacts)
data = transform_data(data)
# split data into train and test sets
data = data.coalesce(1)
data_train, data_test = split_train_test(data, train_split=0.8)
# split data, protected characteristics and label
data_train, protected_train, data_test, protected_test = slice_protected(
data_train, data_test
)
data_train, label_train, data_test, label_test = slice_label(
data_train, data_test
)
# save all sets
save(data_train, "data_train", overwrite=True)
save(protected_train, "protected_train", overwrite=True)
save(label_train, "label_train", overwrite=True)
save(data_test, "data_test", overwrite=True)
save(protected_test, "protected_test", overwrite=True)
save(label_test, "label_test", overwrite=True)
if __name__ == "__main__":
valid_params = ["JOB_NAME", "TempDir"] # base params
valid_params += ["GLUE_DATABASE", "GLUE_REGION"] # custom params
args = getResolvedOptions(sys.argv, valid_params)
glueContext, spark = setup_contexts()
# create and initialize job
job = create_job(glueContext, args)
# define global constants
GLUE_DATABASE = args["GLUE_DATABASE"]
GLUE_REGION = args["GLUE_REGION"]
S3_BUCKET, S3_PREFIX = get_database_location(GLUE_DATABASE, GLUE_REGION)
GLUE_TEMP = args["TempDir"]
# define job in `main` and then commit.
main()
job.commit()