EDA(Exploratory Data Analysis) 노트북을 사용하여 예측 모델에 대한 웹 기반 데이터 탐색

!pip install colorama
import chart_studio.plotly as py
import plotly.graph_objs as go
from plotly.offline import iplot
from scipy import stats
import numpy as np
import warnings
warnings.filterwarnings('ignore')
from scipy.stats import pearsonr
import matplotlib.pyplot as plt
from scipy.stats import pearsonr
import pandas as pd
import math
import re
import seaborn as sns
from datetime import datetime
import colorama
from colorama import Fore, Style
pd.set_option('display.max_columns', None)
pd.set_option('display.max_rows', None)
pd.set_option('display.width', 1000)
pd.set_option('display.expand_frame_repr', False)
pd.set_option('display.max_colwidth', -1)

qs_connect()

target_table = "table_name"
target_table_id = "table_id"

target_table = "cross_industry_demo_midvalues"
target_table_id = "5f7c40ef488de5194ba0157a"

%%read_sql -c QS_CONNECTION
SELECT distinct Year(timestamp) as Year, Month(timestamp) as Month, count(distinct DAY(timestamp)) as Count_days, min(DAY(timestamp)) as First_date, max(DAY(timestamp)) as Last_date, count(timestamp) as Count_hits
from {target_table}
group by Month(timestamp), Year(timestamp)
order by Year, Month;

target_year = "2020" ## The target year
target_month = "02" ## The target month
target_day = "(01,02,03)" ## The target days

from platform_sdk.dataset_reader import DatasetReader
from datetime import date
dataset_reader = DatasetReader(PLATFORM_SDK_CLIENT_CONTEXT, dataset_id=target_table_id)
# If you do not see any data or would like to expand the default date range, change the following query
Table = dataset_reader.limit(5).read()

print("\nNumber of columns:",len(Table.columns))

Table.head(5)

ColumnNames_Types = pd.DataFrame(Table.dtypes)
ColumnNames_Types = ColumnNames_Types.reset_index()
ColumnNames_Types.columns = ["Column_Name", "Data_Type"]
ColumnNames_Types

trace = go.Bar(
    x = query_2_df['Hour'],
    y = query_2_df['Count'],
    name = "Activity Count"
)

layout = go.Layout(
    title = 'Activity Count by Hour of Day',
    width = 1200,
    height = 600,
    xaxis = dict(title = 'Hour of Day'),
    yaxis = dict(title = 'Count')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

trace = go.Bar(
    x = query_4_df['Page_Name'],
    y = query_4_df['Page_Views'],
    name = "Page Views"
)

layout = go.Layout(
    title = 'Top Ten Viewed Pages For a Given Day',
    width = 1000,
    height = 600,
    xaxis = dict(title = 'Page_Name'),
    yaxis = dict(title = 'Page_Views')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

trace = go.Bar(
    x = query_6_df['state_city'],
    y = query_6_df['Count'],
    name = "Activity by City"
)

layout = go.Layout(
    title = 'Top Ten Cities by User Activity',
    width = 1200,
    height = 600,
    xaxis = dict(title = 'City'),
    yaxis = dict(title = 'Count')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

trace = go.Bar(
    x = "SKU-" + query_7_df['Product_SKU'],
    y = query_7_df['Total_Product_Views'],
    name = "Product View"
)

layout = go.Layout(
    title = 'Top Ten Viewed Products',
    width = 1200,
    height = 600,
    xaxis = dict(title = 'SKU'),
    yaxis = dict(title = 'Product View Count')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

goal = "commerce.`order`.purchaseID" #### prediction variable
goal_column_type = "numerical" #### choose either "categorical" or "numerical"
prediction_window_day_start = "2020-01-01" #### YYYY-MM-DD
prediction_window_day_end = "2020-01-31" #### YYYY-MM-DD
analysis_period_day_start = "2020-02-01" #### YYYY-MM-DD
analysis_period_day_end = "2020-02-28" #### YYYY-MM-DD

### If the goal is a categorical goal then select threshold for the defining category and creating bins. 0 is no order placed, and 1 is at least one order placed:
threshold = 1

Data = pd.merge(agg_data,target_df, on='ID',how='left')
Data['TARGET'].fillna(0, inplace=True)

Data.shape

Data.head(5)

print("Count of unique profiles:", (len(Data)))

Missing = pd.DataFrame(round(Data.isnull().sum()*100/len(Data),2))
Missing.columns =['Percentage_missing_values']
Missing['Features'] = Missing.index

trace = go.Bar(
    x = Missing['Features'],
    y = Missing['Percentage_missing_values'],
    name = "Percentage_missing_values")

layout = go.Layout(
    title = 'Missing values',
    width = 1200,
    height = 600,
    xaxis = dict(title = 'Features'),
    yaxis = dict(title = 'Percentage of missing values')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

TARGET = Data.TARGET

Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
Data_numerical.drop(['TARGET'],axis = 1,inplace = True)
Data_numerical1 = Data_numerical

for i in range(0,len(Data_numerical1.columns)):
    Q1 = Data_numerical1.iloc[:,i].quantile(0.25)
    Q3 = Data_numerical1.iloc[:,i].quantile(0.75)
    IQR = Q3 - Q1
    Data_numerical1.iloc[:,i] = np.where(Data_numerical1.iloc[:,i]<(Q1 - 1.5 * IQR),np.nan, np.where(Data_numerical1.iloc[:,i]>(Q3 + 1.5 * IQR),
                                                                                                    np.nan,Data_numerical1.iloc[:,i]))

Outlier = pd.DataFrame(round(Data_numerical1.isnull().sum()*100/len(Data),2))
Outlier.columns =['Percentage_outliers']
Outlier['Features'] = Outlier.index

trace = go.Bar(
    x = Outlier['Features'],
    y = Outlier['Percentage_outliers'],
    name = "Percentage_outlier")

layout = go.Layout(
    title = 'Outliers',
    width = 1200,
    height = 600,
    xaxis = dict(title = 'Features'),
    yaxis = dict(title = 'Percentage of outliers')
)

fig = go.Figure(data = [trace], layout = layout)
iplot(fig)

Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
distribution = pd.DataFrame([Data_numerical.count(),Data_numerical.mean(),Data_numerical.quantile(0), Data_numerical.quantile(0.01),
                             Data_numerical.quantile(0.05),Data_numerical.quantile(0.25), Data_numerical.quantile(0.5),
                        Data_numerical.quantile(0.75),  Data_numerical.quantile(0.95),Data_numerical.quantile(0.99), Data_numerical.max()])
distribution = distribution.T
distribution.columns = ['Count', 'Mean', 'Min', '1st_perc','5th_perc','25th_perc', '50th_perc','75th_perc','95th_perc','99th_perc','Max']
distribution

A = sns.palplot(sns.color_palette("Blues"))

for column in Data_numerical.columns[0:]:
    plt.figure(figsize=(5, 4))
    plt.ticklabel_format(style='plain', axis='y')
    sns.distplot(Data_numerical[column], color = A, kde=False, bins=6, hist_kws={'alpha': 0.4});

Data_categorical = Data.select_dtypes(include='object')
Data_categorical.drop(['ID'], axis = 1, inplace = True, errors = 'ignore')

for column in Data_categorical.columns[0:]:
    if (len(Data_categorical[column].value_counts())>10):
        plt.figure(figsize=(12, 8))
        sns.countplot(x=column, data = Data_categorical, order = Data_categorical[column].value_counts().iloc[:10].index, palette="Set2");
    else:
        plt.figure(figsize=(12, 8))
        sns.countplot(x=column, data = Data_categorical, palette="Set2");

for col in Data.columns:
    if len(Data[col].unique()) == 1:
        if col == 'TARGET':
            print(Fore.RED + '\033[1m' + 'WARNING: TARGET HAS A SINGLE UNIQUE VALUE, ANY BIVARIATE ANALYSIS (NEXT STEP IN THIS NOTEBOOK) OR PREDICTION WILL BE MEANINGLESS' + Fore.RESET + '\x1b[21m')
        elif col == 'ID':
            print(Fore.RED + '\033[1m' + 'WARNING: THERE IS ONLY ONE PROFILE IN THE DATA, ANY BIVARIATE ANALYSIS (NEXT STEP IN THIS NOTEBOOK) OR PREDICTION WILL BE MEANINGLESS' + Fore.RESET + '\x1b[21m')
        else:
            print('Dropped column:',col)
            Data.drop(col,inplace=True,axis=1)

#### Select only numerical data
Data_numerical = Data.select_dtypes(include=['float64', 'int64'])

#### For columns that contain days we impute max days of history for null values, for rest all we impute 0

# Imputing days with max days of history
Days_cols = [col for col in Data_numerical.columns if 'DAYS_' in col]
d1 = datetime.strptime(analysis_period_day_start, "%Y-%m-%d")
d2 = datetime.strptime(analysis_period_day_end, "%Y-%m-%d")
A = abs((d2 - d1).days)

for column in Days_cols:
    Data[column].fillna(A, inplace=True)

# Imputing 0
Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
Missing_numerical = Data_numerical.columns[Data_numerical.isnull().any()].tolist()

for column in Missing_numerical:
    Data[column].fillna(0, inplace=True)

#### Correct for missing values in categorical columns (Replace with mode)
Data_categorical = Data.select_dtypes(include='object')
Missing_cat = Data_categorical.columns[Data_categorical.isnull().any()].tolist()
for column in Missing_cat:
    Data[column].fillna(Data[column].mode()[0], inplace=True)

if len(Data) == 1:
    print(Fore.RED + '\033[1m' + 'THERE IS ONLY ONE PROFILE IN THE DATA, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE AT LEAST ONE MORE PROFILE TO DO BIVARIATE ANALYSIS')
elif len(Data['TARGET'].unique()) == 1:
    print(Fore.RED + '\033[1m' + 'TARGET HAS A SINGLE UNIQUE VALUE, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE PROFILES WITH ATLEAST ONE DIFFERENT VALUE OF TARGET TO DO BIVARIATE ANALYSIS')
else:
    if (goal_column_type == "categorical"):
        TARGET_categorical = pd.DataFrame(np.where(TARGET>=threshold,"1","0"))
        TARGET_categorical.rename(columns={TARGET_categorical.columns[0]: "TARGET_categorical" }, inplace = True)
        Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
        Data_numerical.drop(['TARGET'],inplace=True,axis=1)
        Data_numerical = pd.concat([Data_numerical, TARGET_categorical.astype(int)], axis = 1)
        ncols_for_charts = len(Data_numerical.columns)-1
        nrows_for_charts = math.ceil(ncols_for_charts/4)
        fig, axes = plt.subplots(nrows=nrows_for_charts, ncols=4, figsize=(18, 15))
        for idx, feat in enumerate(Data_numerical.columns[:-1]):
            ax = axes[int(idx // 4), idx % 4]
            sns.boxplot(x='TARGET_categorical', y=feat, data=Data_numerical, ax=ax)
            ax.set_xlabel('')
            ax.set_ylabel(feat)
            fig.tight_layout();
    else:
        Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
        TARGET = pd.DataFrame(Data_numerical.TARGET)
        Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
        Data_numerical.drop(['TARGET'],inplace=True,axis=1)
        Data_numerical = pd.concat([Data_numerical, TARGET.astype(int)], axis = 1)
        for i in Data_numerical.columns[:-1]:
            sns.pairplot(x_vars=i, y_vars=['TARGET'], data=Data_numerical, height = 4)
        f, ax = plt.subplots(figsize = (10,8))
        corr = Data_numerical.corr()

if len(Data) == 1:
    print(Fore.RED + '\033[1m' + 'THERE IS ONLY ONE PROFILE IN THE DATA, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE AT LEAST ONE MORE PROFILE TO DO BIVARIATE ANALYSIS')
elif len(Data['TARGET'].unique()) == 1:
    print(Fore.RED + '\033[1m' + 'TARGET HAS A SINGLE UNIQUE VALUE, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE PROFILES WITH ATLEAST ONE DIFFERENT VALUE OF TARGET TO DO BIVARIATE ANALYSIS')
else:
    if (goal_column_type == "categorical"):
        TARGET_categorical = pd.DataFrame(np.where(TARGET>=threshold,"1","0"))
        TARGET_categorical.rename(columns={TARGET_categorical.columns[0]: "TARGET_categorical" }, inplace = True)
        Data_categorical = Data.select_dtypes(include='object')
        Data_categorical.drop(["ID"], axis =1, inplace = True)
        Cat_columns = Data_categorical
        Data_categorical = pd.concat([TARGET_categorical,Data_categorical], axis =1)
        for column in Cat_columns.columns:
            A = Data_categorical[column].value_counts().iloc[:10].index
            Data_categorical1 = Data_categorical[Data_categorical[column].isin(A)]
            plt.figure(figsize=(12, 8))
            sns.countplot(x="TARGET_categorical",hue=column, data = Data_categorical1, palette = 'Blues')
            plt.xlabel("GOAL")
            plt.ylabel("COUNT")
            plt.show();
    else:
        Data_categorical = Data.select_dtypes(include='object')
        Data_categorical.drop(["ID"], axis =1, inplace = True)
        Target = Data.TARGET
        Data_categorical = pd.concat([Data_categorical,Target], axis =1)
        for column in Data_categorical.columns[:-1]:
            A = Data_categorical[column].value_counts().iloc[:10].index
            Data_categorical1 = Data_categorical[Data_categorical[column].isin(A)]
            sns.catplot(x=column, y="TARGET", kind = "boxen", data =Data_categorical1, height=5, aspect=13/5);

if len(Data) == 1:
    print(Fore.RED + '\033[1m' + 'THERE IS ONLY ONE PROFILE IN THE DATA, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE AT LEAST ONE MORE PROFILE TO FIND IMPORTANT VARIABLES')
elif len(Data['TARGET'].unique()) == 1:
    print(Fore.RED + '\033[1m' + 'TARGET HAS A SINGLE UNIQUE VALUE, BIVARIATE ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE PROFILES WITH ATLEAST ONE DIFFERENT VALUE OF TARGET TO FIND IMPORTANT VARIABLES')
else:
    Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
    Correlation = pd.DataFrame(Data_numerical.drop("TARGET", axis=1).apply(lambda x: x.corr(Data_numerical.TARGET)))
    Correlation['Corr_abs'] = abs(Correlation)
    Correlation = Correlation.sort_values(by = 'Corr_abs', ascending = False)
    Imp_features = pd.DataFrame(Correlation.index[0:10])
    Imp_features.rename(columns={0:'Important Feature'}, inplace=True)
    print(Imp_features)

# Proxy for monetary value is TOTAL_ORDER_REVENUE and proxy for frequency is NUMBER_VISITS
if len(Data) == 1:
    print(Fore.RED + '\033[1m' + 'THERE IS ONLY ONE PROFILE IN THE DATA, INSIGHTS ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE AT LEAST ONE MORE PROFILE TO FIND IMPORTANT VARIABLES')
elif len(Data['TARGET'].unique()) == 1:
    print(Fore.RED + '\033[1m' + 'TARGET HAS A SINGLE UNIQUE VALUE, INSIGHTS ANALYSIS IS NOT APPLICABLE, PLEASE INCLUDE PROFILES WITH ATLEAST ONE DIFFERENT VALUE OF TARGET TO FIND IMPORTANT VARIABLES')
else:
    sns.lmplot("DAYS_SINCE_VISIT", "TOTAL_ORDER_REVENUE", Data, hue="TARGET", fit_reg=False);

TARGET = Data.TARGET

Data_numerical = Data.select_dtypes(include=['float64', 'int64'])
Data_numerical.drop(['TARGET'],axis = 1,inplace = True)

for i in range(0,len(Data_numerical.columns)):
    Q1 = Data_numerical.iloc[:,i].quantile(0.25)
    Q3 = Data_numerical.iloc[:,i].quantile(0.75)
    IQR = Q3 - Q1
    Data_numerical.iloc[:,i] = np.where(Data_numerical.iloc[:,i]<(Q1 - 1.5 * IQR), (Q1 - 1.5 * IQR), np.where(Data_numerical.iloc[:,i]>(Q3 + 1.5 * IQR),
                                                                                                     (Q3 + 1.5 * IQR),Data_numerical.iloc[:,i]))
Data_categorical = Data.select_dtypes(include='object')
Data = pd.concat([Data_categorical, Data_numerical, TARGET], axis = 1)