sys.exit("Argument missing - lambda_v is required")
print(
"===================================MF Option Setting==================================="
)
print("\tbinarizing ratings - %s" % binary_rating)
print("\tdata path - %s" % data_path)
print("\tresult path - %s" % res_dir)
print("\tpretrained w2v data path - %s" % pretrain_w2v)
print ("\tdimension: %d\n\tlambda_u: %.4f\n\tlambda_v: %.4f\n\tmax_iter: %d\n\tnum_kernel_per_ws: %d" \
% (dimension, lambda_u, lambda_v, max_iter, num_kernel_per_ws))
print(
"==========================================================================================="
)
R, D_all = data_factory.load(aux_path, binary_rating)
train_user = data_factory.read_rating(data_path + '/train_user.dat',
binary_rating)
train_item = data_factory.read_rating(data_path + '/train_item.dat',
binary_rating)
valid_user = data_factory.read_rating(data_path + '/valid_user.dat',
binary_rating)
test_user = data_factory.read_rating(data_path + '/test_user.dat',
binary_rating)
# for each user, build a query contains user id, ground truth of top_n items, and pre-selected items
print("Making query for each user...")
query_list = []
all_item_set = set(range(R.shape[1]))
for i in range(R.shape[0]):
q = Query(i)
q.extendGroundTruth(valid_user[0][i])
q.extendGroundTruth(test_user[0][i])
if len(q.ground_truth) != 0:
print("===================================ConvMF Option Setting===================================")
print("\taux path - %s" % aux_path)
print("\tdata path - %s" % data_path)
print("\tresult path - %s" % res_dir)
print("\tpretrained w2v data path - %s" % pretrain_w2v)
print("\tdimension: %d\n\tlambda_u: %.4f\n\tlambda_v: %.4f\n\tmax_iter: %d\n\tnum_kernel_per_ws: %d" \
% (dimension, lambda_u, lambda_v, max_iter, num_kernel_per_ws))
print("===========================================================================================")
R, D_all = data_factory.load(aux_path)
CNN_X = D_all['X_sequence']
vocab_size = len(D_all['X_vocab']) + 1
from models import ConvMF
if pretrain_w2v is None:
init_W = None
else:
init_W = data_factory.read_pretrained_word2vec(
pretrain_w2v, D_all['X_vocab'], emb_dim)
train_user = data_factory.read_rating(data_path + '/train_user.dat')
train_item = data_factory.read_rating(data_path + '/train_item.dat')
valid_user = data_factory.read_rating(data_path + '/valid_user.dat')
test_user = data_factory.read_rating(data_path + '/test_user.dat')
ConvMF(max_iter=max_iter, res_dir=res_dir,
lambda_u=lambda_u, lambda_v=lambda_v, dimension=dimension, vocab_size=vocab_size, init_W=init_W, give_item_weight=give_item_weight, CNN_X=CNN_X, emb_dim=emb_dim,
num_kernel_per_ws=num_kernel_per_ws,
train_user=train_user, train_item=train_item, valid_user=valid_user, test_user=test_user, R=R)
from data_manager import Data_Factory
from rating_models import PMF
import pickle
import sys
import os
data_factory = Data_Factory()
data = sys.argv[1]
file = data.split('/')[2].split('.')[0]
test_file = "./test/" + file + "/"
output_file = "./outputs/" + file + "/"
if not os.path.exists(output_file):
os.makedirs(output_file)
binary_rating = False
R = pickle.load(open(test_file + 'ratings.all', 'rb'))
train_user = data_factory.read_rating(test_file + 'train_user.dat', binary_rating)
train_item = data_factory.read_rating(test_file + 'train_item.dat', binary_rating)
valid_user = data_factory.read_rating(test_file + 'valid_user.dat', binary_rating)
test_user = data_factory.read_rating(test_file + 'test_user.dat', binary_rating)
'''PMF'''
PMF(res_dir=output_file, lambda_u=0.1, lambda_v=0.1, train_user=train_user, train_item=train_item, valid_user=valid_user, test_user=test_user, R=R)
'''
import numpy as np
def eval_MAE(R, U, V, TS):
num_user = U.shape[0]
sub_mae = np.zeros(num_user)
TS_count = 0
for i in xrange(num_user):
idx_item = TS[i]
if len(idx_item) == 0:
continue
TS_count = TS_count + len(idx_item)
approx_R_i = U[i].dot(V[idx_item].T) # approx_R[i, idx_item]
R_i = R[i]
sub_mae[i] = np.abs(approx_R_i - R_i).sum()
mae = sub_mae.sum() / TS_count
return mae
from data_manager import Data_Factory
#import numpy as np
data_factory = Data_Factory()
data_path = '/home/daicoolb/CopeData/test_0723_0.8_100k'
#R,D_all,S=data_factory.load(data_path)
#train_user=data_factory.read_rating(data_path+'/train_user.dat')
test_user = data_factory.read_rating(data_path + '/test_user.dat')
U = np.loadtxt(data_path + '/Result/U.dat')
V = np.loadtxt(data_path + '/Result/V.dat')
print "MAE: %.5f \n " % eval_MAE(test_user[1], U, V, test_user[0])
print "===================================ConvMF Option Setting==================================="
print "\taux path - %s" % aux_path
print "\tdata path - %s" % data_path
print "\tresult path - %s" % res_dir
print "\tpretrained w2v data path - %s" % pretrain_w2v
print "\tdimension: %d\n\tlambda_u: %.4f\n\tlambda_v: %.4f\n\tmax_iter: %d\n\tnum_kernel_per_ws: %d" \
% (dimension, lambda_u, lambda_v, max_iter, num_kernel_per_ws)
print "==========================================================================================="
R, D_all = data_factory.load(aux_path)
CNN_X = D_all['X_sequence']
vocab_size = len(D_all['X_vocab']) + 1
from models import ConvMF
if pretrain_w2v is None:
init_W = None
else:
init_W = data_factory.read_pretrained_word2vec(
pretrain_w2v, D_all['X_vocab'], emb_dim)
train_user = data_factory.read_rating(data_path + '/train_user.dat')
train_item = data_factory.read_rating(data_path + '/train_item.dat')
valid_user = data_factory.read_rating(data_path + '/valid_user.dat')
test_user = data_factory.read_rating(data_path + '/test_user.dat')
ConvMF(max_iter=max_iter, res_dir=res_dir,
lambda_u=lambda_u, lambda_v=lambda_v, dimension=dimension, vocab_size=vocab_size, init_W=init_W, give_item_weight=give_item_weight, CNN_X=CNN_X, emb_dim=emb_dim, num_kernel_per_ws=num_kernel_per_ws,
train_user=train_user, train_item=train_item, valid_user=valid_user, test_user=test_user, R=R)
print "\tContent: %s" % (print_helper(content_mode))
if 'cnn' in content_mode:
print "\tnum_kernel_per_ws: %d\n\tpretrained w2v data path - %s" % (
num_kernel_per_ws, pretrain_w2v)
print('\tItem weight %s ' %
('Constant (a=1,b=0,01)'
if not give_item_weight else 'Constant (a=1,b=0,01). And f(n)'))
if 'cnn_cae' in content_mode:
print '\tJoin CNN and CAE outputs method: %s' % (
'Transfer block' if use_transfer_block else 'Concatenation')
print "==========================================================================================="
for f in range(1, fold_num + 1):
train_user = data_factory.read_rating(
os.path.join(data_path, 'fold-{}'.format(f),
'train-fold_{}-users.dat'.format(f)))
train_item = data_factory.read_rating(
os.path.join(data_path, 'fold-{}'.format(f),
'train-fold_{}-items.dat'.format(f)))
# in case of training only on training and test sets
if os.path.exists(
os.path.join(data_path, 'fold-{}'.format(f),
'validation-fold_{}-users.dat'.format(f))):
valid_user = data_factory.read_rating(
os.path.join(data_path, 'fold-{}'.format(f),
'validation-fold_{}-users.dat'.format(f)))
else:
valid_user = None
test_user = data_factory.read_rating(
os.path.join(data_path, 'fold-{}'.format(f),