Learning Objectives
In this notebook, we will prepare the babyweight dataset for model development and training to predict the weight of a baby before it is born. We will use BigQuery to perform data augmentation and preprocessing which will be used for AutoML Tables, BigQuery ML, and Keras models trained on Cloud AI Platform.
In this lab, we will set up the environment, create the project dataset, preprocess and augment natality dataset, create the train and eval tables in BigQuery, and export data from BigQuery to GCS in CSV format.
Each learning objective will correspond to a #TODO in this student lab notebook -- try to complete this notebook first and then review the solution notebook.
Check that the Google BigQuery library is installed and if not, install it.
In [ ]:
!sudo chown -R jupyter:jupyter /home/jupyter/training-data-analyst
In [1]:
%%bash
sudo pip freeze | grep google-cloud-bigquery==1.6.1 || \
sudo pip install google-cloud-bigquery==1.6.1
Import necessary libraries.
In [2]:
import os
from google.cloud import bigquery
Set environment variables so that we can use them throughout the entire lab. We will be using our project name for our bucket, so you only need to change your project and region.
In [ ]:
%%bash
export PROJECT=$(gcloud config list project --format "value(core.project)")
echo "Your current GCP Project Name is: "$PROJECT
In [13]:
# TODO: Change environment variables
PROJECT = "cloud-training-demos" # REPLACE WITH YOUR PROJECT NAME
BUCKET = "BUCKET" # REPLACE WITH YOUR PROJECT NAME, DEFAULT BUCKET WILL BE PROJECT ID
REGION = "us-central1" # REPLACE WITH YOUR BUCKET REGION e.g. us-central1
# Do not change these
os.environ["BUCKET"] = PROJECT if BUCKET == "BUCKET" else BUCKET # DEFAULT BUCKET WILL BE PROJECT ID
os.environ["REGION"] = REGION
if PROJECT == "cloud-training-demos":
print("Don't forget to update your PROJECT name! Currently:", PROJECT)
Our dataset is hosted in BigQuery. The CDC's Natality data has details on US births from 1969 to 2008 and is a publically available dataset, meaning anyone with a GCP account has access. Click here to access the dataset.
The natality dataset is relatively large at almost 138 million rows and 31 columns, but simple to understand. weight_pounds is the target, the continuous value we’ll train a model to predict.
In [ ]:
%%bash
# Create a BigQuery dataset for babyweight if it doesn't exist
datasetexists=$(bq ls -d | grep -w babyweight)
if [ -n "$datasetexists" ]; then
echo -e "BigQuery dataset already exists, let's not recreate it."
else
echo "Creating BigQuery dataset titled: babyweight"
bq --location=US mk --dataset \
--description "Babyweight" \
$PROJECT:babyweight
echo "Here are your current datasets:"
bq ls
fi
## Create GCS bucket if it doesn't exist already...
exists=$(gsutil ls -d | grep -w gs://${BUCKET}/)
if [ -n "$exists" ]; then
echo -e "Bucket exists, let's not recreate it."
else
echo "Creating a new GCS bucket."
gsutil mb -l ${REGION} gs://${BUCKET}
echo "Here are your current buckets:"
gsutil ls
fi
Since there is already a publicly available dataset, we can simply create the training and evaluation data tables using this raw input data. First we are going to create a subset of the data limiting our columns to weight_pounds, is_male, mother_age, plurality, and gestation_weeks as well as some simple filtering and a column to hash on for repeatable splitting.
We have some preprocessing and filtering we would like to do to get our data in the right format for training.
Preprocessing:
is_male from BOOL to STRINGplurality from INTEGER to STRING where [1, 2, 3, 4, 5] becomes ["Single(1)", "Twins(2)", "Triplets(3)", "Quadruplets(4)", "Quintuplets(5)"]hashcolumn hashing on year and monthFiltering:
20000000
In [6]:
%%bigquery
CREATE OR REPLACE TABLE
babyweight.babyweight_data AS
SELECT
weight_pounds,
CAST(is_male AS STRING) AS is_male,
mother_age,
CASE
WHEN plurality = 1 THEN "Single(1)"
WHEN plurality = 2 THEN "Twins(2)"
WHEN plurality = 3 THEN "Triplets(3)"
WHEN plurality = 4 THEN "Quadruplets(4)"
WHEN plurality = 5 THEN "Quintuplets(5)"
END AS plurality,
gestation_weeks,
FARM_FINGERPRINT(
CONCAT(
CAST(year AS STRING),
CAST(month AS STRING)
)
) AS hashmonth
FROM
publicdata.samples.natality
WHERE
year > 2000
AND weight_pounds > 0
AND mother_age > 0
AND plurality > 0
AND gestation_weeks > 0
Out[6]:
In [7]:
%%bigquery
CREATE OR REPLACE TABLE
babyweight.babyweight_augmented_data AS
SELECT
weight_pounds,
is_male,
mother_age,
plurality,
gestation_weeks,
hashmonth
FROM
babyweight.babyweight_data
UNION ALL
SELECT
weight_pounds,
"Unknown" AS is_male,
mother_age,
CASE
WHEN plurality = "Single(1)" THEN plurality
ELSE "Multiple(2+)"
END AS plurality,
gestation_weeks,
hashmonth
FROM
babyweight.babyweight_data
Out[7]:
In [8]:
%%bigquery
CREATE OR REPLACE TABLE
babyweight.babyweight_data_train AS
SELECT
weight_pounds,
is_male,
mother_age,
plurality,
gestation_weeks
FROM
babyweight.babyweight_augmented_data
WHERE
ABS(MOD(hashmonth, 4)) < 3
Out[8]:
In [9]:
%%bigquery
CREATE OR REPLACE TABLE
babyweight.babyweight_data_eval AS
SELECT
weight_pounds,
is_male,
mother_age,
plurality,
gestation_weeks
FROM
babyweight.babyweight_augmented_data
WHERE
ABS(MOD(hashmonth, 4)) = 3
Out[9]:
In [10]:
%%bigquery
-- LIMIT 0 is a free query; this allows us to check that the table exists.
SELECT * FROM babyweight.babyweight_data_train
LIMIT 0
Out[10]:
In [11]:
%%bigquery
-- LIMIT 0 is a free query; this allows us to check that the table exists.
SELECT * FROM babyweight.babyweight_data_eval
LIMIT 0
Out[11]:
Use BigQuery Python API to export our train and eval tables to Google Cloud Storage in the CSV format to be used later for TensorFlow/Keras training. We'll want to use the dataset we've been using above as well as repeat the process for both training and evaluation data.
In [14]:
# Construct a BigQuery client object.
client = bigquery.Client()
dataset_name = "babyweight"
# Create dataset reference object
dataset_ref = client.dataset(
dataset_id=dataset_name, project=client.project)
# Export both train and eval tables
for step in ["train", "eval"]:
destination_uri = os.path.join(
"gs://", BUCKET, dataset_name, "data", "{}*.csv".format(step))
table_name = "babyweight_data_{}".format(step)
table_ref = dataset_ref.table(table_name)
extract_job = client.extract_table(
table_ref,
destination_uri,
# Location must match that of the source table.
location="US",
) # API request
extract_job.result() # Waits for job to complete.
print("Exported {}:{}.{} to {}".format(
client.project, dataset_name, table_name, destination_uri))
In [15]:
%%bash
gsutil ls gs://${BUCKET}/babyweight/data/*.csv
In [18]:
%%bash
gsutil cat gs://${BUCKET}/babyweight/data/train000000000000.csv | head -5
In [19]:
%%bash
gsutil cat gs://${BUCKET}/babyweight/data/eval000000000000.csv | head -5
Copyright 2019 Google Inc. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License
In [ ]: