import pandas as pd
from sklearn.metrics.pairwise import cosine_similarity

# Read data source Excel files.
df1 = pd.read_excel('Online_Retail.xlsx')
df1a = df1.dropna(subset=['CustomerID'])

# Create CustomerID vs Item (Purchased Items, by StockCode) matrix by pivot table function.
CustomerID_Item_matrix = df1a.pivot_table(
    index='CustomerID', 
    columns='StockCode', 
    values='Quantity',
    aggfunc='sum'
)

# Update illustration of the matrix, 1 to represent customer have purchased item, 0 to represent customer haven't purchased.
CustomerID_Item_matrix = CustomerID_Item_matrix.applymap(lambda x: 1 if x > 0 else 0)

# Create User to User similarity matrix. 
user_to_user_similarity_matrix = pd.DataFrame(
    cosine_similarity(CustomerID_Item_matrix)
)

# Update index to corresponding CustomerID.
user_to_user_similarity_matrix.columns = CustomerID_Item_matrix.index
user_to_user_similarity_matrix['CustomerID'] = CustomerID_Item_matrix.index
user_to_user_similarity_matrix = user_to_user_similarity_matrix.set_index('CustomerID')

# Display CustomerID (12702) purchased items.
items_purchased_by_X = set(CustomerID_Item_matrix.loc[12702.0].iloc[
    CustomerID_Item_matrix.loc[12702.0].to_numpy().nonzero()].index)

# Display CustomerID (14608) purchased items.
items_purchased_by_Y = set(CustomerID_Item_matrix.loc[14608.0].iloc[
    CustomerID_Item_matrix.loc[14608.0].to_numpy().nonzero()].index)

# Find out items which purchased by X (12702) but not yet purchased by Y (14608).
items_to_recommend_to_Y = items_purchased_by_X - items_purchased_by_Y

# Display the list of items recommended for Y (14608) with item Description.
print(df1a.loc[
    df1a['StockCode'].isin(items_to_recommend_to_Y), 
    ['StockCode', 'Description']
].drop_duplicates().set_index('StockCode'))