# Import libraries
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from google.cloud import bigquery

sns.set()

# Launch BigQuery client from the JSON credentials file
bq_client = bigquery.Client.from_service_account_json('../../bigquery_creds.json')

# Helper function for queries to be returned as a Dataframe
def bquery(sql):
    return bq_client.query(sql).to_dataframe()

# Sessions and page views for 10 days
sql = """
SELECT
    PARSE_DATE("%Y%m%d", date) AS date,
    COUNT(DISTINCT fullVisitorId) AS visitors,
    SUM(totals.visits) AS sessions,
    SUM(totals.pageviews) AS pageviews,
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
GROUP BY 1
ORDER BY 1
"""
bquery(sql)

# Traffic KPIs broken down by device and acquisition channel
sql = """
SELECT
    device.deviceCategory AS device,
    channelGrouping AS channel_grouping,
    SUM(totals.visits) AS sessions,
    SUM(totals.pageviews) AS pageviews
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
GROUP BY 1,2
ORDER BY 1,2
"""
bquery(sql)

# Pages per session, average sessions duration (in seconds) and bounce rate
sql = """
SELECT
    PARSE_DATE("%Y%m%d", date) AS date,
    SUM(totals.visits) AS sessions,
    SUM(totals.pageviews)/SUM(totals.visits) AS page_per_session,
    SUM(totals.timeOnSite)/SUM(totals.visits) AS avg_session_duration,
    SUM(totals.bounces)/SUM(totals.visits) AS bounce_rate
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
GROUP BY 1
ORDER BY 1
"""
bquery(sql)

# Transactions and revenue
sql = """
SELECT
    PARSE_DATE("%Y%m%d", date) AS date,
    SUM(totals.visits) AS sessions,
    SUM(totals.transactions) AS transactions,
    SUM(totals.transactions)/SUM(totals.visits) AS conversion_rate,
    SUM(totals.transactionRevenue)/1e6 AS revenue,
    SUM(totals.transactionRevenue)/1e6/SUM(totals.transactions) AS avg_purchase_value
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
GROUP BY 1
ORDER BY 1
"""
bquery(sql)

# Compute if a goal has been reached in each session
sql = """
WITH sub1 AS (
    SELECT
        PARSE_DATE("%Y%m%d", date) AS date,
        CONCAT(fullVisitorId, CAST(visitId AS STRING)) AS session_id,
        -- Goal: browsed to page /registersuccess.html
        MAX(IF(hits.page.pagePath = '/registersuccess.html', 1, 0)) AS goal_destination,
        -- Goal: session duration >= 5 min
        MAX(IF(totals.timeOnSite >= 300, 1, 0)) AS goal_duration,
        -- Goal: pages per session > 10
        MAX(IF(totals.pageviews > 10, 1, 0)) AS goal_pageviews
    FROM
        `bigquery-public-data.google_analytics_sample.ga_sessions_*`,
        UNNEST(hits) AS hits
    WHERE
        _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
        AND totals.visits > 0
    GROUP BY date, session_id
)

SELECT
    date,
    COUNT(session_id) AS sessions,
    SUM(goal_destination) AS goals_destination,
    SUM(goal_duration) AS goals_duration,
    SUM(goal_pageviews) AS goals_pageviews
FROM sub1
GROUP BY date
ORDER BY date
"""
bquery(sql)

# Get number of session before a transaction, and conversion rates by session position
sql = """
WITH sub1 AS (
    SELECT
        fullVisitorId,
        CONCAT(fullVisitorId, CAST(visitId AS STRING)) AS session_id,
        visitStartTime,
        MAX(IF(totals.transactions > 0, 1, 0)) AS transaction_session,
        RANK() OVER (PARTITION BY fullVisitorId ORDER BY visitStartTime) AS session_number
    FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
    WHERE
        _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
    GROUP BY fullVisitorId, session_id, visitStartTime
)

SELECT
    session_number,
    COUNT(session_id) AS sessions,
    SUM(transaction_session) AS transactions,
    SUM(transaction_session)/COUNT(session_id) AS conversion_rate
FROM sub1
GROUP BY session_number
HAVING transactions > 0
ORDER BY session_number
"""
bquery(sql)

# Get time lag (in days) between first session of a user and a converting session
sql = """
WITH sub1 AS (
    SELECT
        fullVisitorId,
        MIN(visitStartTime) OVER (PARTITION BY fullVisitorId ORDER BY visitStartTime) AS first_session_time,
        MAX(IF(totals.transactions > 0, visitStartTime, 0)) AS transaction_session_time
    FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
    WHERE
        _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
    GROUP BY fullVisitorId, visitStartTime
)

SELECT
    FLOOR((transaction_session_time - first_session_time)/3600/24) AS lag_days,
    COUNT(*) AS transactions
FROM sub1
WHERE transaction_session_time > 0
GROUP BY lag_days
ORDER BY lag_days
"""
bquery(sql)

# Differentiate between SEA keywords containing 'google' vs others
sql = """
SELECT
    CASE
        WHEN REGEXP_CONTAINS(LOWER(trafficSource.keyword), r"google") AND channelGrouping = 'Paid Search' THEN 'Paid Search Brand'
        WHEN channelGrouping = 'Paid Search' THEN 'Paid Search Non Brand'
        ELSE 'Other channels'
    END AS channel,
    SUM(totals.visits) AS sessions,
    SUM(totals.transactions) AS transactions,
    SUM(totals.transactions)/SUM(totals.visits) AS conversion_rate,
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
GROUP BY 1
ORDER BY 1
"""
bquery(sql)

# Differentiate True Direct in channels
sql = """
SELECT
    channelGrouping AS channel,
    trafficSource.isTrueDirect IS NOT NULL AS true_direct,
    SUM(totals.visits) AS sessions,
    SUM(totals.transactions) AS transactions,
    ROUND(100*SUM(totals.transactions)/SUM(totals.visits), 2) AS conversion_rate,
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170731'
GROUP BY 1,2
ORDER BY 1,2
"""
bquery(sql)

# Conversion rate by weekday and hour
sql = """
SELECT
    EXTRACT(DAYOFWEEK FROM PARSE_DATE("%Y%m%d", date)) AS weekday,
    EXTRACT(HOUR FROM PARSE_TIMESTAMP("%s", CAST(visitStartTime AS STRING))) AS hour,
    100*SUM(totals.transactions)/SUM(totals.visits) AS conv_rate
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`
WHERE
    _TABLE_SUFFIX BETWEEN '20170701' AND '20170731'
GROUP BY 1,2
"""
slots = bquery(sql)

# Plot as a heatmap
fig, ax = plt.subplots(1, 1, figsize=(8,8))
sns.heatmap(
    slots.pivot(index='hour', columns='weekday', values='conv_rate').fillna(0),
    cmap="YlGnBu", annot=True, fmt='.2f', cbar=False);

# Number of pages viewed in a session before reaching a Goal (destination page)
sql = """
WITH sub1 AS (
    SELECT
        CONCAT(fullVisitorId, CAST(visitId AS STRING)) AS session_id,
        -- Window function to compute rank of each Page viewed in a session
        RANK() OVER (PARTITION BY fullVisitorId, visitId ORDER BY hits.hitNumber) AS page_number,
        hits.page.pagePath AS page
    FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`,
    UNNEST(hits) AS hits
    WHERE
        _TABLE_SUFFIX BETWEEN '20170701' AND '20170710'
        -- Filter on Pages only, we don't want all Hits types
        AND hits.type = 'PAGE'
    GROUP BY fullVisitorId, visitId, hits.hitNumber, page
)

SELECT
    page_number,
    COUNT(*) AS goals
FROM sub1
-- Filter on the destination page associated with a Goal
WHERE page = '/registersuccess.html'
GROUP BY 1
ORDER BY 1
"""
goals_pages = bquery(sql)

# Plot distribution
(
    goals_pages
    .set_index('page_number')
    .reindex(list(range(0, max(goals_pages['page_number'])+1)))
    .plot(kind='bar', figsize=(10,6))
);