collaborative filtering recommender systems example

import numpy as np

NaN = np.nan

import pandas as pd

# The ratings by users (U)
df = pd.DataFrame(np.array([[8,NaN,10,NaN,10],
                           [NaN,2,NaN,3,3],
                           [2,8,NaN,4,1],
                           [8,NaN,NaN,2,NaN],
                           [NaN,7,3,NaN,NaN],
                           [5,NaN,8,1,NaN]]),
                  columns = ['U1', 'U2', 'U3', 'U4', 'UA'])

# These are items
df.rows = ['A', 'B', 'C', 'D', 'E', 'F']


Z = df.sub(df.mean(), axis=1)
# Z = df.add(df.mean() + df, axis=1)
Z


for i in range(len(Z.columns)-1):
  Z['Z'+str(i)+'xZa']=[x*y for x, y  in zip(Z.loc[:,'U'+str(i+1)], Z.loc[:,'UA'])]


w =  Z.sum(axis=0)[5:9]
w= w.values.tolist()
x = w+['']+w
Z.loc['Similarity(W)', :] = x
print("W",w)
print("meam:", m[4])
d = 1/np.sum(w)
print("d:",d)
print("================")
for i in range(3,6):
  s  = 0
  for j in range(0,4):
    a = Z.iloc[i,j]
    b = w[j]
    if (not np.isnan(a) and not np.isnan(b)):
      s += a*b
 
  Z.iloc[i,4] = np.round(m[4] + d*s,2)

from IPython.display import display, HTML

Z=Z.replace(np.nan, '', regex=True)
# Z=Z.round(1)
display(Z)