Skillz Analysis¶
In [1]:
import pandas as pd
from pandas.plotting import register_matplotlib_converters
import matplotlib.pyplot as plt
import seaborn as sns
from pprint import pprint as pp
from pathlib import Path
import matplotlib.dates as mdates
sns.set()
register_matplotlib_converters()
# %matplotlib inline
pandas options
In [2]:
pd.set_option('max_columns', 200)
pd.set_option('max_rows', 300)
pd.set_option('display.expand_frame_repr', True)
Data
In [3]:
data_dir_path = Path.cwd()
skillz_data = data_dir_path / 'Skillz Analytics H2H Homework Data_v0.4.xlsx'
Pandas DataFrame¶
In [4]:
skillz_df = pd.read_excel(skillz_data, '1. Aggregate')
Data Exploration¶
In [5]:
skillz_df.head()
Out[5]:
In [6]:
skillz_df.info()
In [7]:
skillz_df.set_index('date', inplace=True)
In [18]:
skillz_df.index = skillz_df.index.to_period('D')
7-day Centered Rolling Mean¶
In [8]:
numeric_columns = skillz_df.columns[1:]
In [9]:
skillz_7d_control = skillz_df[numeric_columns][skillz_df.experiment_group == 'Control'].rolling(7, center=True).mean()
skillz_7d_test = skillz_df[numeric_columns][skillz_df.experiment_group == 'Test'].rolling(7, center=True).mean()
In [24]:
skillz_7d_control.head(10)
Out[24]:
In [25]:
skillz_7d_test.head(10)
Out[25]:
Weekly Mean¶
In [12]:
skillz_weekly_control = skillz_df[numeric_columns][skillz_df.experiment_group == 'Control'].resample('W').mean()
skillz_weekly_test = skillz_df[numeric_columns][skillz_df.experiment_group == 'Test'].resample('W').mean()
In [20]:
skillz_weekly_control
Out[20]:
In [21]:
skillz_weekly_test
Out[21]:
Plots¶
Daily Counts¶
In [15]:
df_columns = skillz_df.columns[1:]
In [19]:
fix, axes = plt.subplots(8, 1, figsize=(15, 22), sharex=True)
for name, ax in zip(df_columns, axes):
sns.barplot(data=skillz_df, x=skillz_df.index, y=name, ax=ax, hue='experiment_group')
plt.xticks(rotation=90)
Weekly Mean Resample¶
In [16]:
weeks = mdates.WeekdayLocator()
In [17]:
fix, axes = plt.subplots(8, 1, figsize=(15, 60), sharex=True)
for name, ax in zip(df_columns, axes):
ax.plot(skillz_df[name][skillz_df.experiment_group == 'Control'], marker='.', linestyle='-', linewidth=0.5, label='Control: Daily')
ax.plot(skillz_weekly_control[name], marker='o', markersize=8, linestyle='-', label='Control: Weekly Mean Resample')
ax.plot(skillz_7d_control[name], marker='.', markersize=5, linestyle='-', label='Control: 7-Day Rolling Mean')
ax.plot(skillz_df[name][skillz_df.experiment_group == 'Test'], marker='.', linestyle='-', linewidth=0.5, label='Test: Daily')
ax.plot(skillz_weekly_test[name], marker='o', markersize=8, linestyle='-', label='Test: Weekly Mean Resample')
ax.plot(skillz_7d_test[name], marker='.', markersize=5, linestyle='-', label='Test: 7-Day Rolling Mean')
plt.xticks(rotation=90)
ax.set_ylabel(name)
ax.xaxis.set_major_locator(weeks)
ax.legend()
entry_fees vs. cash_daily_active_users¶
In [34]:
plt.subplots(figsize=(7, 7))
sns.scatterplot(skillz_df.cash_daily_active_users, skillz_df.entry_fees, hue=skillz_df.experiment_group)
plt.show()
cash_games vs. entry_fees¶
In [35]:
plt.subplots(figsize=(7, 7))
sns.scatterplot(skillz_df.entry_fees, skillz_df.cash_games, hue=skillz_df.experiment_group)
plt.show()
cash_games vs. deposits¶
In [36]:
plt.subplots(figsize=(7, 7))
sns.scatterplot(skillz_df.deposits, skillz_df.cash_games, hue=skillz_df.experiment_group)
plt.show()
deposits vs. new_depositors¶
In [37]:
plt.subplots(figsize=(7, 7))
sns.scatterplot(skillz_df.new_depositors, skillz_df.deposits, hue=skillz_df.experiment_group)
plt.show()
z_games vs. z_daily_active_users¶
In [38]:
plt.subplots(figsize=(7, 7))
sns.scatterplot(skillz_df.z_daily_active_users, skillz_df.z_games, hue=skillz_df.experiment_group)
plt.show()
Questions¶
The provided data is inadequate and associated questions aren't relavant for fraud analytics
- What additional data would you have liked to see in this analysis?
- Individual account data would be required for fraud detection
- Tagged data to implement supervised learning models
In [ ]:
In [ ]: