Google Analytics のデータを python で アソシエーション分析 する

!python3 -m pip install pandas

!python3 -m pip install pyfpgrowth

import pandas as pd
df = pd.read_table("GA_dataset.tsv")
# client_id と日付の昇順でソート
df = df.sort_values(by=["ga:dimension8", "Date"], ascending=True)
# データ件数が多く後続の計算に時間を擁するのでデータを減らす
df = df[1:3000]

transactions = []
strs = []
previous_client_id = ""
for idx in df.index:
    elems = df.ix[idx]
    client_id = elems["ga:dimension8"]
    if(previous_client_id != ""):
        if (previous_client_id != client_id):
            transactions.append(strs)
            strs = []
            strs.extend(((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',').split(','))
        else:
            strs.extend(((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',').split(','))
    else:
        strs.extend(((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',').split(','))
    previous_client_id = client_id
if strs:
    transactions.append(strs)

import pyfpgrowth
patterns = pyfpgrowth.find_frequent_patterns(transactions, 3)
rules = pyfpgrowth.generate_association_rules(patterns, 0.7)

patterns

rules.get(('/blog/Block-multiple-requests-of-Wicket-Ajax/', '/blog/apache-wicketでrestapiを使う/','/blog/category/wicket/?page=2', '/blog/category/wicket/?page=2', '/blog/category/wicket/?page=2', '/blog/wicket-scriptタグ-を-body-閉じタグの直前に出力する/'))

!python3 -m pip install orange3
!python3 -m pip install orange3-associate

lines = []
line = ""
previous_client_id = ""
for idx in df.index:
    elems = df.ix[idx]
    client_id = elems["ga:dimension8"]
    if(previous_client_id != ""):
        if (previous_client_id != client_id):
            lines.append(line)
            line = ""
            line = line + ((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',')
        else:
            line = line + ((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',')
    else:
        line = line + ((elems["Page"] + ",") * elems["Pageviews"]).rstrip(',')
    previous_client_id = client_id
if (line != ""):
    lines.append(line)

file = open("dataset.basket","w")
for line in lines:
    file.write(line + "\n")

import Orange
from orangecontrib.associate.fpgrowth import *
tbl = Orange.data.Table('dataset.basket')

/translate_c?depth=1&hl=en&prev=search&rurl=translate.google.com&sl=ja&sp=nmt4&u=https://www.monotalk.xyz/blog/google-apps-script-で電子署名をする/&usg=ALkJrhhsVbHe1cSGBqRWm5dF2OEu936DqQ

lines = []
line = ""
previous_client_id = ""
for idx in df.index:
    elems = df.ix[idx]
    client_id = elems["ga:dimension8"]
    if(previous_client_id != ""):
        if (previous_client_id != client_id):
            lines.append(line.rstrip(',').replace("?","").replace("=","").replace("&",""))
            line = ""
            line = line + ((elems["Page"] + ",") * elems["Pageviews"])
        else:
            line = line + ((elems["Page"] + ",") * elems["Pageviews"])
    else:
        line = line + ((elems["Page"] + ",") * elems["Pageviews"])
    previous_client_id = client_id
if (line != ""):
    lines.append(line.rstrip(',').replace("?","").replace("=","").replace("&",""))

file = open("dataset.basket","w")
for line in lines:
    file.write(line + "\n")

import Orange
from orangecontrib.associate.fpgrowth import *
tbl = Orange.data.Table('dataset.basket')

def decode_onehot(d):
    items = OneHot.decode(d, tbl, mapping)
    # ContinuousVariable の name 値を取得
    return list(map(lambda v: v[1].name, items))

X, mapping = OneHot.encode(tbl)
# 2 は組み合わせごとの発生回数、2件以上のものを取得する
itemsets = dict(frequent_itemsets(X, 2))

# アソシエーションルールの抽出
# 信頼度(確信度)が0.7以上のものを抽出する
rules = association_rules(itemsets, 0.7)
for P, Q, support, confidence in rules:
    lhs = decode_onehot(P)
    rhs = decode_onehot(Q)
    print(f"lhs = {lhs}, rhs = {rhs}, support = {support}, confidence = {confidence}")

def decode_onehot(d):
    items = OneHot.decode(d, tbl, mapping)
    # ContinuousVariable の name 値を取得
    return list(map(lambda v: v[1].name, items))

X, mapping = OneHot.encode(tbl)
# .001 は組み合わせごとの発生確率、0.1%以上のものを取得する
# 結果が出力されるのが、0.1%以下でのみ出力されました。
itemsets = dict(frequent_itemsets(X, .001))

# アソシエーションルールの抽出
# 信頼度(確信度)が0.7以上のものを抽出する
rules = association_rules(itemsets, 0.7)
for P, Q, support, confidence in rules:
    lhs = decode_onehot(P)
    rhs = decode_onehot(Q)
    print(f"lhs = {lhs}, rhs = {rhs}, support = {support}, confidence = {confidence}")

stats = rules_stats(rules, itemsets, len(X))
for s in sorted(stats, key = lambda x: x[6], reverse = True):

    lhs = decode_onehot(s[0])
    rhs = decode_onehot(s[1])

    support = s[2]
    confidence = s[3]
    lift = s[6]

    print(f"lhs = {lhs}, rhs = {rhs}, support = {support}, confidence = {confidence}, lift = {lift}")