def predict(self, itemList, Y, R, n=10, normalize=True):
assert self.X is not None
assert self.Theta is not None
# make predictions by multiplying X and Theta
self.predictions = self.X.dot(self.Theta.T)[:, 0]
if normalize:
# normalize ratings
Ynorm, Ymean = utils.normalizeRatings(Y, R)
self.predictions = self.predictions + Ymean
assert n < len(self.predictions)
# get the indices ix that would sort predictions in a descending order
# by predictions[ix]
self.ix = np.argsort(self.predictions)[::-1]
self.top = []
i = 0
while i < n:
j = self.ix[i]
if self.ratings[j] == 0: # only recommend unrated items
self.top.append((np.rint(self.predictions[j]), itemList[j]))
i += 1
def train(self,
Y,
R,
lambd=10,
normalize=True,
num_features=10,
maxiter=100):
print("\nTraining model...")
# reshape ratings into a column vector
ratings = np.reshape(self.ratings, (self.ratings.size, 1))
# add the new user's ratings to Y and R
Y = np.concatenate((ratings, Y), axis=1)
R = np.concatenate(((ratings != 0), R), axis=1)
Y_in = Y
if normalize:
# normalize ratings
Ynorm, Ymean = utils.normalizeRatings(Y, R)
Y_in = Ynorm
# useful values
num_users = Y.shape[1]
num_items = Y.shape[0]
# set initial parameters (Theta, X)
X = np.random.randn(num_items, num_features)
Theta = np.random.randn(num_users, num_features)
initial_parameters = np.concatenate((X, Theta))
initial_parameters = np.reshape(initial_parameters,
initial_parameters.size)
# set functions for the cost and the gradient
f = lambda t: utils.costFunc(t, Y_in, R, num_users, num_items,
num_features, lambd)[0]
fprime = lambda t: utils.costFunc(t, Y_in, R, num_users, num_items,
num_features, lambd)[1]
# minimize gradient with fmincg
def callback(x):
if self.count % 10 == 0:
print("iteration {}\tloss {}".format(self.count, f(x)))
self.count += 1
theta = fmin_cg(f=f,
x0=initial_parameters,
fprime=fprime,
maxiter=maxiter,
callback=callback)
# unfold theta back into X and Theta
self.X = np.reshape(theta[0:num_items * num_features],
(num_items, num_features))
self.Theta = np.reshape(theta[num_items * num_features:],
(num_users, num_features))
def predict(new_ratings, _id):
Y, R = loadData()
Y = np.hstack([Y, new_ratings[:, None]])
R = np.hstack([R, (new_ratings[:, None] > 0)])
print("Ratings added!")
# Normalize Ratings
Ynorm, Ymean = utils.normalizeRatings(Y, R)
# Useful Values
num_movies, num_users = Y.shape
num_features = 10
# Set Initial Parameters (Theta, X)
X = np.random.randn(num_movies, num_features)
Theta = np.random.randn(num_users, num_features)
initial_parameters = np.concatenate([X.ravel(), Theta.ravel()])
# Set options for scipy.optimize.minimize
options = {'maxiter': 100}
# Set Regularization
lambda_ = 10
res = optimize.minimize(lambda x: cofiCostFunc(
x, Ynorm, R, num_users, num_movies, num_features, lambda_),
initial_parameters,
method='TNC',
jac=True,
options=options)
theta = res.x
# Unfold the returned theta back into U and W
X = theta[:num_movies * num_features].reshape(num_movies, num_features)
Theta = theta[num_movies * num_features:].reshape(num_users, num_features)
print('Recommender system learning completed.')
p = np.dot(X, Theta.T)
print(p.shape)
my_predictions = p[:, _id] + Ymean
movieList = utils.loadMovieList()
ix = np.argsort(my_predictions)[::-1]
return ix[20:], my_predictions
# Load data
data = loadmat(os.path.join('Data', 'ex8_movies.mat'))
Y, R = data['Y'], data['R']
# Y is a 1682x943 matrix, containing ratings (1-5) of 1682 movies by
# 943 users
# R is a 1682x943 matrix, where R(i,j) = 1 if and only if user j gave a
# rating to movie i
# Add our own ratings to the data matrix
Y = np.hstack([my_ratings[:, None], Y])
R = np.hstack([(my_ratings > 0)[:, None], R])
# Normalize Ratings
Ynorm, Ymean = utils.normalizeRatings(Y, R)
# Useful Values
num_movies, num_users = Y.shape
num_features = 10
# Set Initial Parameters (Theta, X)
X = np.random.randn(num_movies, num_features)
Theta = np.random.randn(num_users, num_features)
initial_parameters = np.concatenate([X.ravel(), Theta.ravel()])
# Set options for scipy.optimize.minimize
options = {'maxiter': 100}
# Set Regularization
