### retrain with best parameter
print('Best parameters:',
[best_momentum, best_reg, best_alpha, best_B, best_F, best_epoch])
# pickle
pickle.dump(best_W, open(fileName, 'wb'))
### END ###
# This part you can write on your own
# you could plot the evolution of the training and validation RMSEs for example
print('Predicting')
# log training and validating error
print(
'RSME on training',
lib.rmse(
trStats["ratings"],
rbm.predict(trStats["movies"], trStats["users"], best_W, training)))
print(
'RSME on validation',
lib.rmse(
vlStats["ratings"],
rbm.predict(vlStats["movies"], vlStats["users"], best_W, training)))
predictedRatings = np.array([
rbm.predictForUser(user, best_W, training) for user in trStats["u_users"]
])
print(np.max(predictedRatings), np.min(predictedRatings))
# Output the 300 x 97 txt
print('Saving new result')
np.savetxt("predictedRatings.txt", predictedRatings)
# We predict over the training set
tr_r_hat = rbm.predict(trStats["movies"],
trStats["users"],
W,
training,
b=biasOne)
trRMSE = lib.rmse(trStats["ratings"], tr_r_hat)
# We predict over the validation set
vl_r_hat = rbm.predict(vlStats["movies"],
vlStats["users"],
W,
training,
b=biasOne)
vlRMSE = lib.rmse(vlStats["ratings"], vl_r_hat)
print "### EPOCH %d ###" % epoch
print "Training loss = %f" % trRMSE
print "Validation loss = %f" % vlRMSE
# break
### END ###
# This part you can write on your own
# you could plot the evolution of the training and validation RMSEs for example
predictedRatings = np.array([
rbm.predictForUser(user, W, training, b=biasOne)
for user in trStats["u_users"]
])
# print(predictedRatings.shape)
np.savetxt("predictedRatingsExp.txt", predictedRatings)
print('Predicting')
# log training and validating error
print(
'RSME on training',
lib.rmse(
trStats["ratings"],
rbm.predict(trStats["movies"],
trStats["users"],
best_W,
training,
predictType="mean")))
print(
'RSME on validation',
lib.rmse(
vlStats["ratings"],
rbm.predict(vlStats["movies"],
vlStats["users"],
best_W,
training,
predictType="mean")))
predictedRatings = np.array([
rbm.predictForUser(user, best_W, training, predictType="mean")
for user in trStats["u_users"]
])
print(np.max(predictedRatings), np.min(predictedRatings))
# Output the 300 x 97 txt
print('Saving new result')
np.savetxt("SUccess+v4.txt", predictedRatings)
count = 0
for F in range_F:
for epsilon in range_Epsilon:
for B in range_B:
for weightcost in range_weightcost:
for momentum in range_momentum:
count += 1
print("====== ", round(count/float(total)*100,2),"% =====")
print("F: ",F)
print("epsilon: ",epsilon)
print("B: ", B)
print("weightcost", weightcost)
print("momentum: ", momentum)
results = run_RBM(F, epochs, epsilon, B, weightcost, momentum, f, mode)
if results[1] < finalRMSE:
finalRMSE = results[1]
finalW = results[0]
if mode == 'single':
finalW = run_RBM(F, epochs, epsilon, B, weightcost, momentum, f, mode)[0]
f.close()
### END ###
# This part you can write on your own
# you could plot the evolution of the training and validation RMSEs for example
predictedRatings = np.array([rbm.predictForUser(user, finalW, training) for user in trStats["u_users"]])
np.savetxt("predictedRatings.txt", predictedRatings)
for B in range_B:
for weightcost in range_weightcost:
for momentum in range_momentum:
count += 1
print("====== ", round(count / float(total) * 100, 2),
"% =====")
print("F: ", F)
print("epsilon: ", epsilon)
print("B: ", B)
print("weightcost", weightcost)
print("momentum: ", momentum)
results = run_RBM(F, epochs, epsilon, B, weightcost,
momentum, f, mode)
if results[1] < finalRMSE:
finalRMSE = results[1]
finalW = results[0]
if mode == 'single':
finalW = run_RBM(F, epochs, epsilon, B, weightcost, momentum, f, mode)[0]
f.close()
### END ###
# This part you can write on your own
# you could plot the evolution of the training and validation RMSEs for example
predictedRatings = np.array([
rbm.predictForUser(user, finalW, training) for user in trStats["u_users"]
])
np.savetxt("predictedRatings.txt", predictedRatings)
