def getA(training):
A = np.zeros((trStats["n_ratings"], trStats["n_movies"] + trStats["n_users"]))
no_movies = trStats["n_movies"]
data = lib.getTrainingData()
for i, rating in enumerate(data):
movie, user, r = rating
A[i][movie] = 1
A[i][no_movies+user] = 1
print (A)
return A
def getA(training):
# A = np.zeros((trStats["n_ratings"], trStats["n_movies"] + trStats["n_users"]))
# for i in range(0, trStats["n_ratings"] - 1):
# A[i][training[i][0]] = training[i][2]
# A[i][training[i] + trStats["n_movies"] - 1] = training[i][2]
# A = np.zeros(trStats["n_users"], trStats["n_movies"])
# for i in range(0, trStats["n_ratings"]):
# A[training[i][1]][training[i][0]] = A[training[i][2]]
# return A
A = np.zeros(
(trStats["n_ratings"], trStats["n_movies"] + trStats["n_users"]))
no_movies = trStats["n_movies"]
data = lib.getTrainingData()
for i, rating in enumerate(data):
movie, user, r = rating
A[i][movie] = 1
A[i][no_movies + user] = 1
return A
import numpy as np
import rbm
import projectLib as lib
import pickle
training = lib.getTrainingData()
validation = lib.getValidationData()
# You could also try with the chapter 4 data
# training = lib.getChapter4Data()
trStats = lib.getUsefulStats(training)
vlStats = lib.getUsefulStats(validation)
K = 5
# SET PARAMETERS HERE!!!
# number of hidden units
# F = 10
epochs = 30
# gradientLearningRate = 0.1
# file for past trained data
fileName = 'best_W'
try:
# try to load existing trained weight, if any
best_W = pickle.load(open(fileName, 'rb'))
except:
# otherwise, train/tune
## Parameter to tune
import numpy as np
import rbm
import projectLib as lib
import pickle
training = lib.getTrainingData()
validation = lib.getValidationData()
# You could also try with the chapter 4 data
# training = lib.getChapter4Data()
trStats = lib.getUsefulStats(training)
vlStats = lib.getUsefulStats(validation)
K = 5
# mode = 'batch'
mode = 'single'
finalW = np.zeros([5,5,5])
finalRMSE = 100
def run_RBM(F, epochs, epsilon, B, weightcost, momentum, f, mode):
# Initialise all our arrays
W = rbm.getInitialWeights(trStats["n_movies"], F, K)
bestW = np.zeros(W.shape)
bestRMSE = 100
posprods = np.zeros(W.shape)
negprods = np.zeros(W.shape)
grad = np.zeros(W.shape)
