Skip to content

Instantly share code, notes, and snippets.

@nkt1546789
Last active August 29, 2016 07:59
Show Gist options
  • Star 1 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save nkt1546789/05cdecbc0a7533ef3027 to your computer and use it in GitHub Desktop.
Save nkt1546789/05cdecbc0a7533ef3027 to your computer and use it in GitHub Desktop.
Weighted Matrix Factorization for Explicit Feedback
def WMF(R,k=2,alpha=0.1):
m,n=R.shape
X=rand(m,k)
Y=rand(n,k)
for itr in xrange(10000):
Xold=X.copy()
Yold=Y.copy()
for i in xrange(m):
ri=R[i]
C=(ri!=0)*1
if sum(C)==0:
continue
X[i]=dot(pinv(dot(Y.T,c_[C]*Y)+alpha*identity(k)),dot(Y.T,C*ri))
for j in xrange(n):
rj=R[:,j]
C=(rj!=0)*1
if sum(C)==0:
continue
Y[j]=dot(pinv(dot(X.T,c_[C]*X)+alpha*identity(k)),dot(X.T,C*rj))
if norm(X-Xold)+norm(Y-Yold)<1e-6:
break
return X,Y
# R=X.dot(Y.T)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment