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

# In[2]:


import datetime
import numpy as np
import pandas as pd
print(pd.__version__)


# In[156]:


start = datetime.datetime(2020, 1, 1)
end = datetime.datetime(2022, 1, 1)
# Create a DatetimeIndex with freq='Q'
# the 'Q' indicates the last day of each quarter
ts = pd.date_range(start, end, freq='M')


# In[157]:


df = pd.DataFrame(index=ts.to_period('M'), data={'Quarter':ts.to_period('Q')})


# In[158]:


# make a sine wave as data
df['Data'] = np.sin(np.linspace(0,10, len(ts)))


# In[159]:


df['pct'] = df.groupby(by='Quarter')['Data'].transform(lambda x: x/sum(x))
df['cumpct'] = df.groupby(by='Quarter')['Data'].transform(lambda x: np.cumsum(x)/sum(x))
df['qtrTot'] = df.groupby(by='Quarter')['Data'].transform(lambda x: x.sum())


# In[160]:


df.head()


# In[161]:


# We can do a shift of 3 months for our cumulative percent for the previous quarter
df['prev_qtr_pct'] = df.cumpct.shift(3)
# then divide it to apply the prediction
df['prev_qtr_linearity'] = df['Data'] / df['prev_qtr_pct']


# In[162]:


# calculate the delta
df['delta'] = (df['prev_qtr_linearity']/df['qtrTot']) -1
# filter on month1/month2
scores = df.query('cumpct!=1 & delta.notna() & delta <=10000').delta


# In[163]:


# MAPE:
np.abs(scores).mean()


# In[164]:


np.abs(scores).plot(kind='bar')


# In[67]:


df.groupby(by='Quarter').sum()['Data'].plot()


# In[ ]: