Explore and visualize data in BigQuery from within JupyterLab

Explore and visualize data in BigQuery from within JupyterLab

This page shows you some examples of how to explore and visualize data that is stored in BigQuery from within the JupyterLab interface of your Vertex AI Workbench instance.

Before you begin

If you haven't already, create a Vertex AI Workbench instance.

Required roles

To ensure that your instance's service account has the necessary permissions to query data in BigQuery, ask your administrator to grant your instance's service account the Service Usage Consumer (roles/serviceusage.serviceUsageConsumer) IAM role on the project. For more information about granting roles, see Manage access to projects, folders, and organizations.

Your administrator might also be able to give your instance's service account the required permissions through custom roles or other predefined roles.

Open JupyterLab

1. In the Google Cloud console, go to the Instances page.

   Go to Instances

2. Next to your Vertex AI Workbench instance's name, click Open JupyterLab.

   Your Vertex AI Workbench instance opens JupyterLab.

Read data from BigQuery

In the next two sections, you read data from BigQuery that you will use to visualize later. These steps are identical to those in Query data in BigQuery from within JupyterLab, so if you've completed them already, you can skip to Get a summary of data in a BigQuery table.

Query data by using the %%bigquery magic command

In this section, you write SQL directly in notebook cells and read data from BigQuery into the Python notebook.

Magic commands that use a single or double percentage character (% or %%) let you use minimal syntax to interact with BigQuery within the notebook. The BigQuery client library for Python is automatically installed in a Vertex AI Workbench instance. Behind the scenes, the %%bigquery magic command uses the BigQuery client library for Python to run the given query, convert the results to a pandas DataFrame, optionally save the results to a variable, and then display the results.

Note: As of version 1.26.0 of the google-cloud-bigquery Python package, the BigQuery Storage API is used by default to download results from the %%bigquery magics.

1. To open a notebook file, select File > New > Notebook.

2. In the Select Kernel dialog, select Python 3, and then click Select.

   Your new IPYNB file opens.

3. To get the number of regions by country in the international_top_terms dataset, enter the following statement:

   %%bigquery
   SELECT
     country_code,
     country_name,
     COUNT(DISTINCT region_code) AS num_regions
   FROM
     `bigquery-public-data.google_trends.international_top_terms`
   WHERE
     refresh_date = DATE_SUB(CURRENT_DATE, INTERVAL 1 DAY)
   GROUP BY
     country_code,
     country_name
   ORDER BY
     num_regions DESC;
   

4. Click  Run cell.

   The output is similar to the following:

   Query complete after 0.07s: 100%|██████████| 4/4 [00:00<00:00, 1440.60query/s]
   Downloading: 100%|██████████| 41/41 [00:02<00:00, 20.21rows/s]
   country_code      country_name    num_regions
   0   TR  Turkey         81
   1   TH  Thailand       77
   2   VN  Vietnam        63
   3   JP  Japan          47
   4   RO  Romania        42
   5   NG  Nigeria        37
   6   IN  India          36
   7   ID  Indonesia      34
   8   CO  Colombia       33
   9   MX  Mexico         32
   10  BR  Brazil         27
   11  EG  Egypt          27
   12  UA  Ukraine        27
   13  CH  Switzerland    26
   14  AR  Argentina      24
   15  FR  France         22
   16  SE  Sweden         21
   17  HU  Hungary        20
   18  IT  Italy          20
   19  PT  Portugal       20
   20  NO  Norway         19
   21  FI  Finland        18
   22  NZ  New Zealand    17
   23  PH  Philippines    17
   ...
   

   Note: Your results might differ from what is above as the google_trends dataset being queried is refreshed with new data on an ongoing basis.

5. In the next cell (below the output from the previous cell), enter the following command to run the same query, but this time save the results to a new pandas DataFrame that's named regions_by_country. You provide that name by using an argument with the %%bigquery magic command.

   %%bigquery regions_by_country
   SELECT
     country_code,
     country_name,
     COUNT(DISTINCT region_code) AS num_regions
   FROM
     `bigquery-public-data.google_trends.international_top_terms`
   WHERE
     refresh_date = DATE_SUB(CURRENT_DATE, INTERVAL 1 DAY)
   GROUP BY
     country_code, country_name
   ORDER BY
     num_regions DESC;
   

   Note: For more information about available arguments for the %%bigquery command, see the client library magics documentation.

6. Click  Run cell.

7. In the next cell, enter the following command to look at the first few rows of the query results that you just read in:

   regions_by_country.head()
   

8. Click  Run cell.

   The pandas DataFrame regions_by_country is ready to plot.

Query data by using the BigQuery client library directly

In this section, you use the BigQuery client library for Python directly to read data into the Python notebook.

The client library gives you more control over your queries and lets you use more complex configurations for queries and jobs. The library's integrations with pandas enable you to combine the power of declarative SQL with imperative code (Python) to help you analyze, visualize, and transform your data.

Note: You can use a number of Python data analysis, data wrangling, and visualization libraries, such as numpy, pandas, matplotlib, and many others. Several of these libraries are built on top of a DataFrame object.

1. In the next cell, enter the following Python code to import the BigQuery client library for Python and initialize a client:

   from google.cloud import bigquery
   
   client = bigquery.Client()
   

   The BigQuery client is used to send and receive messages from the BigQuery API.

2. Click  Run cell.

3. In the next cell, enter the following code to retrieve the percentage of daily top terms in the US top_terms that overlap across time by number of days apart. The idea here is to look at each day's top terms and see what percentage of them overlap with the top terms from the day before, 2 days prior, 3 days prior, and so on (for all pairs of dates over about a month span).

   sql = """
   WITH
     TopTermsByDate AS (
       SELECT DISTINCT refresh_date AS date, term
       FROM `bigquery-public-data.google_trends.top_terms`
     ),
     DistinctDates AS (
       SELECT DISTINCT date
       FROM TopTermsByDate
     )
   SELECT
     DATE_DIFF(Dates2.date, Date1Terms.date, DAY)
       AS days_apart,
     COUNT(DISTINCT (Dates2.date || Date1Terms.date))
       AS num_date_pairs,
     COUNT(Date1Terms.term) AS num_date1_terms,
     SUM(IF(Date2Terms.term IS NOT NULL, 1, 0))
       AS overlap_terms,
     SAFE_DIVIDE(
       SUM(IF(Date2Terms.term IS NOT NULL, 1, 0)),
       COUNT(Date1Terms.term)
       ) AS pct_overlap_terms
   FROM
     TopTermsByDate AS Date1Terms
   CROSS JOIN
     DistinctDates AS Dates2
   LEFT JOIN
     TopTermsByDate AS Date2Terms
     ON
       Dates2.date = Date2Terms.date
       AND Date1Terms.term = Date2Terms.term
   WHERE
     Date1Terms.date <= Dates2.date
   GROUP BY
     days_apart
   
   ORDER BY
     days_apart;
   """
   pct_overlap_terms_by_days_apart = client.query(sql).to_dataframe()
   
   pct_overlap_terms_by_days_apart.head()
   

   The SQL being used is encapsulated in a Python string and then passed to the query() method to run a query. The to_dataframe method waits for the query to finish and downloads the results to a pandas DataFrame by using the BigQuery Storage API.

4. Click  Run cell.

   The first few rows of query results appear below the code cell.

      days_apart   num_date_pairs  num_date1_terms overlap_terms   pct_overlap_terms
    0          0             32               800            800            1.000000
    1          1             31               775            203            0.261935
    2          2             30               750             73            0.097333
    3          3             29               725             31            0.042759
    4          4             28               700             23            0.032857
   

   Note: Your results might differ from what is above as the google_trends dataset being queried is refreshed with new data on an ongoing basis.

For more information about using BigQuery client libraries, see the quickstart Using client libraries.

Get a summary of data in a BigQuery table

In this section, you use a notebook shortcut to get summary statistics and visualizations for all fields of a BigQuery table. This can be a fast way to profile your data before exploring further.

The BigQuery client library provides a magic command, %bigquery_stats, that you can call with a specific table name to provide an overview of the table and detailed statistics on each of the table's columns.

1. In the next cell, enter the following code to run that analysis on the US top_terms table:

   %bigquery_stats bigquery-public-data.google_trends.top_terms
   

2. Click  Run cell.

   After running for some time, an image appears with various statistics on each of the 7 variables in the top_terms table. The following image shows part of some example output:

   

Note: Your results might differ from what is above as the google_trends dataset being queried is refreshed with new data on an ongoing basis.

Visualize BigQuery data

In this section, you use plotting capabilities to visualize the results from the queries that you previously ran in your Jupyter notebook.

1. In the next cell, enter the following code to use the pandas DataFrame.plot() method to create a bar chart that visualizes the results of the query that returns the number of regions by country:

   regions_by_country.plot(kind="bar", x="country_name", y="num_regions", figsize=(15, 10))
   

2. Click  Run cell.

   The chart is similar to the following:

   

3. In the next cell, enter the following code to use the pandas DataFrame.plot() method to create a scatter plot that visualizes the results from the query for the percentage of overlap in the top search terms by days apart:

   pct_overlap_terms_by_days_apart.plot(
     kind="scatter",
     x="days_apart",
     y="pct_overlap_terms",
     s=len(pct_overlap_terms_by_days_apart["num_date_pairs"]) * 20,
     figsize=(15, 10)
     )
   

4. Click  Run cell.

   The chart is similar to the following. The size of each point reflects the number of date pairs that are that many days apart in the data. For example, there are more pairs that are 1 day apart than 30 days apart because the top search terms are surfaced daily over about a month's time.

   

For more information about data visualization, see the pandas documentation.

What's next

To see an example of exploring and visualizing BigQuery data as part of a comprehensive workflow in Vertex AI Workbench, run the "Interactive exploratory analysis of BigQuery data in a notebook" Jupyter notebook in one of the following environments:

Open in Colab | Open in Colab Enterprise | Open in Vertex AI Workbench user-managed notebooks | View on GitHub