Insights_VIZ.py

#!/usr/bin/env python
# coding: utf-8

# ### Package

# import pandas as pd
# import numpy as np
# import numpy as np, pandas as pd
# import seaborn as sns
# import matplotlib.pyplot as plt
# from sklearn.preprocessing import LabelEncoder

# ### Data 

# `old_cust_df`  is a **merged** dataframe containing *customer demographic*, *customer transactions* and *customer address* data-sets
# 
# `new_cust_df`  is a **merged** dataframe containing old and new *customer demographic* and *customer address* data-sets

# old_cust_df = pd.read_csv("Customer_Details_Old_Final.csv")
# new_cust_df = pd.read_csv("Customer_DF_Final.csv")
# old_cust_df.drop(columns="Unnamed: 0", inplace=True)
# new_cust_df.drop(columns="Unnamed: 0", inplace=True)
# print("Dimension of {}: {}".format("old_cust_df", old_cust_df.shape))

# ### Preprocess 

# new_cust_df["state"] = new_cust_df["state"].replace(["NSW", "QLD", "VIC"],["New South Wales", "Queensland", "Victoria"])
# old_cust_df["state"] = old_cust_df["state"].replace(["NSW", "QLD", "VIC"],["New South Wales", "Queensland", "Victoria"])

# set(old_cust_df['state'])

# set(new_cust_df["state"].to_list())

# ### Plots

# ####  Correlation Matrix, an overall view


mat = new_cust_df.corr()
mas = np.triu(np.ones_like(mat, dtype=bool))
col = sns.color_palette("gnuplot2")
sns.heatmap(mat, mask=mas, cmap=col, vmax=.3, center=0,
            square=True, linewidths=.5, cbar_kws={"shrink": .5})

# **Observations**: 
# 
# * It's seen that `Tenure` and `Age` are highly correlated, i.e `Tenure` increases along with `Age`
# * The `Past 3 years purchases` feature has strong correlation with `Property valuation`


def box(x,y,color,xlab,ylab):
    grph = sns.boxplot(x=x, y=y, data=new_cust_df, color = color)
    grph.set(xlabel=xlab, ylabel = ylab)


# #### Bike purchase in 3 years vs. Wealth Segment Distribution 


box("wealth_segment", "past_3_years_bike_related_purchases", "darkturquoise", 
    "Wealth Class", "Bike purchases - 3 Years")

# **Observations**: 
# 
# * All the three wealth classes have a `Median` 3 years buying almost around `≈50`
# * All the three classes seems to have a `Normal Distribution`
# * `50%`of the data in the distribution for all the thee classes lie between `≈27 - ≈75` purchases 
# * The distributions are spreaded with a range of `≈100` 3 years purchases 

# #### Bike purchase in 3 years vs. Gender 

box("gender", "past_3_years_bike_related_purchases", "mediumpurple", "Gender", 
    "Bike purchases - 3 Years")


# **Observations**: 
# 
# * From the plot it is visible that the `Female` category has a greater range than that of the `Male`
# * The distribution of the `Female` is completely `Normal` whereas the `Male` is not symmetric
# * `50%`of the data for the `Female` class lie between `≈27 - ≈75` purchases 

# #### Bike purchase in 3 years vs. Industry Class

# In[15]:

ax = sns.barplot(y="job_industry_category", x="past_3_years_bike_related_purchases", 
                 data=new_cust_df, palette="husl")
ax.set(xlabel="Bike purchases - 3 Years", ylabel = "Job Industry")

# **Observations**: 
# 
# * According to the data, customers who are in `Agriculture` sector have invested more on bikes in past 3 years
# * The customers in `Telecommunications` are the lowest in terms of `Mean` bike purchase in past 3 years 
# * The `Mean` purchases among the `Entertainment` field lies between `0 - 47` 
# * The purchases data in the `Manufacturing` field are more accurately distributed and are more reliable for analysis
# * The `Telecommunications` purchase data is the least reliabe. The distribution is not tight.

# #### Bike purchase in 3 years vs. State 

ax = sns.catplot(y="state",x="past_3_years_bike_related_purchases",data=new_cust_df, palette="Paired")
ax.set(xlabel="Bike purchases - 3 Years", ylabel = "State")


# **Observations**: 
# 
# * The customers from `New South Wales` has the most purchase rate that the remaining states
# * The `Queensland` along with the `Victoria` data are unreliable for analysis due to less data-points
# * There are less data points from `Victoria` hence the distribution is highly spread out with a huge `Error`
# * The data-points from `NSW` are tight enough and highly reliable with the least `Error`

# #### Bike purchase in 3 years vs. Own cars 

ax = sns.catplot(y="owns_car", x="past_3_years_bike_related_purchases", kind="bar", data=new_cust_df)
ax.set(xlabel="Bike purchases - 3 Years", ylabel = "Own cars")


# **Observations**
# 
# * The customers with `No cars` tend to have slightly bigger `Mean` of bikes in 3 years

# *`Point Biserial`* correlation of purchase rate against the categories

encode = LabelEncoder()
new_cust_df["Cars_bool"] = encode.fit_transform(new_cust_df["owns_car"])
new_cust_df_corr = new_cust_df[["past_3_years_bike_related_purchases", "owns_car", "Cars_bool"]]
new_cust_df_corr.rename(columns={"past_3_years_bike_related_purchases": "Bike purchases - 3 Years", 
                                 "Cars_bool": "Own cars"}, inplace = True)

new_cust_df_corr.corr()


# **Observations**
# 
# * From `Point Biserial Corr.` we can see that owning a car and purchase of bikes in 3 years have `Negative correlation`
# * The correlation between Bike purchase in 3 years and customers owning cars is `-0.009`

# ####  Bike purchase in 3 years vs. Product Size (Other)


ax = sns.catplot(y="product_line",x="past_3_years_bike_related_purchases",data=old_cust_df)
ax.set(xlabel="Bike purchases - 3 Years", ylabel = "Product Line")


# **Observations**
# 
# * The `Standard` bikes were the favourite followed by the `Road` line
# * The customers were not that fond of `Mountain` in the given 3 years
# * The distribution for `Touring` bikes are sparse due to less data-points

# By: *`Adipta`*