import pandas as pd
import numpy as np, numpy.random
import random
import matplotlib.pyplot as plt

# variables
final_dataframe_size = 1000 # in this case, these are the 1000 customers who have bought something.

# Generate users
users = [x for x in range (1000, 10000)]
users[:10]

# shuffle the original list
random.shuffle(users)
users[:10]

tran_users = random.sample(users, final_dataframe_size)
print (len(tran_users))
print (tran_users[:10])

pareto_tran_users = random.sample(tran_users, int(.2 * final_dataframe_size))
len(pareto_tran_users)

df = pd.DataFrame({'user': tran_users})
df.head()

def assign_amount(val):
    if val in pareto_tran_users:
        return 400
    return 25
     
df['transaction'] = df.user.apply(assign_amount)

# validate pareto
df.sort_values('transaction', ascending=False).head(int(.2 * final_dataframe_size)).sum()/df.sum()

def sum_to_x(n, x):
    values = [0.0, x] + list(np.random.uniform(low=0.0,high=x,size=n-1))
    values.sort()
    return [values[i+1] - values[i] for i in range(n)]

df2 = df.copy()

df2_p = df2[df2.user.isin(pareto_tran_users)]
df2_p_n = df2[~df2.user.isin(pareto_tran_users)]
# .apply(sum_to_x(20000, 80000))

df2_p['transaction'] = sum_to_x(200, 80000)
df2_p_n['transaction'] = sum_to_x(800, 20000)
df2 = df2_p.append(df2_p_n)

df2

df2.sum()

# validate pareto
df2.sort_values('transaction', ascending=False).head(200).sum().apply(lambda x: '%.3f' % x)

df_users = pd.read_csv('synthetic_users.csv')

df_users.head()

df_items = pd.read_csv('synthetic_items_catalog.csv')

df_items.drop_duplicates(subset=['product'], keep='first', inplace=True)

df_items['id'] = df_items['id'] + 1000

df_items

prices = np.random.normal(50, 20, 836)

df_items['prices'] = prices
df_items['prices'] = np.where(df_items['prices'] < 0 , 2.0, df_items['prices'])
df_items['prices'] = round(df_items['prices'])

df_items.describe()

np.random.normal(0, 1, 10)

## Could not figure out how to generate normal distribution frame, 
## so opting out for uniform distribution

# weights_users = np.random.normal(1000, 1000, len(df_users.id))
weights_items = np.random.normal(1000, 1000, len(df_items.id))

tran_user = random.choices(random.sample(list(df_users.id), 268), k=10000)
tran_items = random.choices(list(df_items.id), weights = weights_items, k=10000)

df = pd.DataFrame.from_dict({'user_id': tran_user, 'item_id': tran_items})\
.set_index('item_id')\
.join(df_items.set_index('id'))


df_ = df.groupby('user_id')[['prices']].sum()

df_.describe()

# validate pareto
# regardless of transaction table, hamro users 200 jana
df_.reset_index(inplace=True)
(df_.sort_values('prices', ascending=False).head(200).sum())/df_.sum()

merged = df.reset_index()\
.set_index('user_id')\
.join(df_users.set_index('id'), how='outer')\
.reset_index()\
.rename(columns={'index':'product_id', 'level_0': 'user_id'})

merged

merged.groupby('user_id')[['prices']].sum().sort_values('prices', ascending=False).head(200).sum()/merged['prices'].sum()

merged.to_csv('merged.csv', index=False)
df_users.to_csv('users.csv', index=False)
df_items.to_csv('items.csv', index=False)

merged

df_users

df_items