def main():
for number in range(1, 1 + datasets_total):
# time = datetime.datetime.now()
train, test, lengths = load_movielens_dataset(number)
learn, validate = split(train, shuffle=False)
(lam, gamma), predictor = optimize_parameters(learn, validate, lengths)
print("Dataset #%i: RMSE = %6.4f" % (number, svd_sgd.rmse(test, predictor)))
print("Parameters: λ1 = %6.4f, λ2 = %6.4f, ɣ1 = %6.4f, ɣ2 = %6.4f" % (lam[0], lam[1], gamma[0], gamma[1]))
def main():
for number in range(1, 1 + datasets_total):
#time = datetime.datetime.now()
train, test, lengths = load_movielens_dataset(number)
learn, validate = split(train, shuffle=False)
(lam, gamma), predictor = optimize_parameters(learn, validate, lengths)
print("Dataset #%i: RMSE = %6.4f" %
(number, svd_sgd.rmse(test, predictor)))
print("Parameters: λ1 = %6.4f, λ2 = %6.4f, ɣ1 = %6.4f, ɣ2 = %6.4f" %
(lam[0], lam[1], gamma[0], gamma[1]))
def optimize_parameters(train, validate, lengths):
rmse_best, params_best, pred_best = 100, ((0, 0), (0, 0)), lambda u, i: 0
for lam1 in 0.1 ** np.arange(0.0, 3.0):
for lam2 in 0.1 ** np.arange(0.0, 3.0):
for gamma1 in 0.33 ** np.arange(5.0, 8.0):
for gamma2 in 0.33 ** np.arange(5.0, 8.0):
# print(lam1, lam2, gamma1, gamma2)
predictor = svd_sgd.learn(train, lengths, (lam1, lam2), (gamma1, gamma2))
rmse = svd_sgd.rmse(validate, predictor)
# print(rmse)
if rmse < rmse_best:
rmse_best = rmse
params_best = ((lam1, lam2), (gamma1, gamma2))
pred_best = predictor
return params_best, pred_best
def optimize_parameters(train, validate, lengths):
rmse_best, params_best, pred_best = 100, ((0, 0), (0, 0)), lambda u, i: 0
for lam1 in 0.1**np.arange(0.0, 3.0):
for lam2 in 0.1**np.arange(0.0, 3.0):
for gamma1 in 0.33**np.arange(5.0, 8.0):
for gamma2 in 0.33**np.arange(5.0, 8.0):
#print(lam1, lam2, gamma1, gamma2)
predictor = svd_sgd.learn(train, lengths, (lam1, lam2),
(gamma1, gamma2))
rmse = svd_sgd.rmse(validate, predictor)
#print(rmse)
if rmse < rmse_best:
rmse_best = rmse
params_best = ((lam1, lam2), (gamma1, gamma2))
pred_best = predictor
return params_best, pred_best