configParser = ConfigParser.RawConfigParser()
configParser.read(configFilePath)
dataset = configParser.get('shared', 'd')
num_users = int(configParser.get('shared', 'num_users'))
dimension = int(configParser.get('shared', 'dimension'))
rec_per_user = int(configParser.get('shared', 'rec_per_user'))
num_votes = int(configParser.get('shared', 'v'))
train_ratio = float(configParser.get('experiments', 'train_ratio'))
res_path = 'YOUR PATH'
graph_prefix = 'graph_d_' + str(dimension)
scores_prefix = 'pr_users_scores_d_' + str(dimension)
if experiment_mode == 'S':
graph_prefix += '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
scores_prefix += '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
graph_file = res_path + get_file_name(graph_prefix, num_users, train_ratio,
'gml')
graph = nx.read_gml(graph_file, destringizer=literal_destringizer)
users = []
items = []
for node in graph.nodes():
if graph.node[node]['type'] == 'user':
users.append(node)
if graph.node[node]['type'] == 'item':
items.append(node)
# ------------ user pagerank --------------
print 'start computing pagerank'
t1 = time.time()
res = compute_pagerank(graph, users, alpha, epsilon)
print 'finished user pagerank', time.time() - t1
configParser.read(configFilePath)
dataset = configParser.get('shared', 'd')
exp_per_rec = int(configParser.get('shared', 'e'))
item_per_pair = int(configParser.get('shared', 'i'))
num_votes = int(configParser.get('shared', 'v'))
dimension = int(configParser.get('shared', 'dimension'))
rec_per_user = int(configParser.get('shared', 'rec_per_user'))
folder = configParser.get('shared', 'folder')
num_users = configParser.get('shared', 'num_users')
feedback_ratio = float(configParser.get('experiments', 'feedback_ratio'))
train_ratio = float(configParser.get('experiments', 'train_ratio'))
test_ratio = float(configParser.get('experiments', 'test_ratio'))
res_path = 'YOUR PATH'
setup_postfix = '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
test_file = res_path + get_file_name('test', num_users, test_ratio)
if test_strategy == "sample":
sample_test_file = res_path + get_file_name('test_sample', num_users,
test_ratio)
users_samples = read_random_recs(sample_test_file)
#rwr_feedback_file = res_path + folder + get_file_name('RWR_simulated_feedback', num_users, feedback_ratio)
graph_file = res_path + get_file_name('graph', num_users, train_ratio,
'gml')
graph = nx.read_gml(graph_file, destringizer=literal_destringizer)
updated_graph_file = res_path + get_file_name(
'graph_d_' + str(dimension) + setup_postfix, num_users, train_ratio,
'gml')
if os.path.exists(updated_graph_file):
updated_graph = nx.read_gml(updated_graph_file,
destringizer=literal_destringizer)
# read the book description
book_desc_file = path + "books_descriptions.txt"
books_desc_text = {}
with open(book_desc_file, 'r') as f_in:
next(f_in)
for line in f_in:
tabs = line.strip().split('\t')
if len(tabs) >= 2:
books_desc_text[tabs[0]] = tabs[1].replace(''', '')
else:
books_desc_text[tabs[0]] = 'No description found.'
# ---------------------------------- phase2 ---------------------------------
if phase == 2:
# read recs and explanations for RWR
users_scores_file = path + get_file_name('pr_users_scores_d_'+str(dimension), num_users, train_ratio)
items_scores_file = path + get_file_name('pr_items_scores_d_'+str(dimension), num_users, train_ratio)
graph_file = path + get_file_name('graph_d_'+str(dimension), num_users, train_ratio, 'gml')
graph = nx.read_gml(graph_file, destringizer=literal_destringizer)
train_data = get_users_data(path + get_file_name('train_d_' + str(dimension), num_users, train_ratio))
rwr_ur = top_k_recs_rwr(users_scores_file, graph, rec_per_user)
users = rwr_ur.keys()
rwr_ure_good = explanation_items_rwr(items_scores_file, graph, rwr_ur, exp_per_rec)
rwr_ure_bad = explanation_items_rwr(items_scores_file, graph, rwr_ur, exp_per_rec, location='bottom')
# merging data {user: {rec: {explanations: {exp: {M: R: S: P: TW: BW: TF: BF:}} M: R:}}}
output_data = {}
for user in users:
output_data[user] = {}
# merge recs
output_data = add_rec_exps(user, output_data, rwr_ure_good, model='R', location='TW')
feature_reduction = 'nmf'
model = 'RWR'
# read features
if model == 'RWR':
feature_file = path + dataset + "-data-" + feature_reduction + "-features-" + str(
dimension) + ".csv"
features = utils.read_features(feature_file, normalized=True)
# ---------------- only items ------------------------
# generate extended training
recs_file = path + 'phase2/' + model + '_rated_recs_phase_2.txt'
main_train_file = path + 'train_users_' + str(
num_users) + '_partition_100.txt'
file_prefix = 'train_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
new_train_file = path + get_file_name(file_prefix, num_users, 1)
lines_1 = []
with open(main_train_file, 'r') as f_in:
for line in f_in:
lines_1.append(line.strip())
lines_2 = []
with open(recs_file, 'r') as f_in:
next(f_in)
for line in f_in:
lines_2.append(line.strip())
with open(new_train_file, 'w') as f_out:
for line in lines_1:
f_out.write(line)
f_out.write('\n')
for line in lines_2:
f_out.write(line)
folder_name = configParser.get('shared', 'folder')
sim_threshold = float(configParser.get('feedback_inc', 'sim_threshold'))
num_vectors = int(configParser.get('feedback_inc', 'num_vectors'))
mode = configParser.get('feedback_inc', 'mode')
beta = float(configParser.get('shared', 'beta'))
feedback_ratio = float(configParser.get('experiments', 'feedback_ratio'))
train_ratio = float(configParser.get('experiments', 'train_ratio'))
num_users = int(configParser.get('shared', 'num_users'))
res_path = 'YOUR PATH'
input_output_path = res_path + folder_name
# build interaction graph
graph_prefix = 'graph_d_' + str(dimension)
if experiment_mode == 'SP':
graph_prefix += '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
graph_file = res_path + get_file_name(graph_prefix, num_users, train_ratio,
'gml')
graph_nx = nx.read_gml(graph_file, destringizer=literal_destringizer)
graph = InteractionGraph()
graph.set_graph(graph_nx)
# read the weight vectors and item festures
weight_file = input_output_path + get_file_name(
'user_weight_vector_learned', num_users, train_ratio)
if exp_loc == 'bottom':
weight_file = input_output_path + get_file_name(
'user_weight_vector_learned_bottom', num_users, train_ratio)
feature_file = res_path + dataset + "-" + feature_reduction + "-features-" + str(
dimension) + ".csv"
all_features = read_features(feature_file, normalized=True)
user_weights = np.genfromtxt(weight_file, delimiter=',')
print 'read features and weights'
exp_per_rec = int(configParser.get('shared', 'e'))
item_per_pair = int(configParser.get('shared', 'i'))
num_votes = int(configParser.get('shared', 'v'))
folder_name = configParser.get('shared', 'folder')
learning_rate = float(configParser.get('update_sim_unconstr', 'learning_rate'))
weight_decay = float(configParser.get('update_sim_unconstr', 'weight_decay'))
n_epoch = int(configParser.get('update_sim_unconstr', 'n_epoch'))
mode = configParser.get('feedback_inc', 'mode')
feedback_ratio = float(configParser.get('experiments', 'feedback_ratio'))
train_ratio = float(configParser.get('experiments', 'train_ratio'))
path = 'YOUR PATH'
res_path = path + folder_name
feature_file = path + dataset + "-" + feature_reduction + "-features-" + str(dimension) + ".csv"
simulated_feedback_file = res_path + get_file_name(model + '_simulated_feedback_d_'+ str(dimension), num_users,
feedback_ratio)
if exp_loc == 'bottom':
simulated_feedback_file = res_path + get_file_name(model + '_simulated_feedback_bottom', num_users,
feedback_ratio)
print 'read bottom'
points_numpy = utils.read_features(feature_file, normalized=True)
item_id_map = {int(points_numpy[i, 0]): i for i in range(points_numpy.shape[0])}
points = torch.from_numpy(points_numpy[:, 1:].T)
print 'started computing cross products'
points_cross_points = torch.matmul(torch.t(points), points)
dimension = points.shape[0]
num_items = points.shape[1]
# read the feedback pairs
users_feedback_pairs = {}
users = []
print 'started reading the pairs'
num_users = int(configParser.get('shared', 'num_users'))
dimension = int(configParser.get('shared', 'dimension'))
rec_per_user = int(configParser.get('shared', 'rec_per_user'))
num_votes = int(configParser.get('shared', 'v'))
train_partition = float(configParser.get('experiments', 'train_ratio'))
sim_threshold = float(configParser.get('feedback_inc', 'sim_threshold'))
beta = float(configParser.get('shared', 'beta'))
sim_file = 'item-item-similarity-' + str(dimension) + '.txt'
train_prefix = 'train'
graph_prefix = 'graph_d_' + str(dimension)
if experiment_mode == 'S': # the graph contains user's feedback on recommendations
train_prefix = 'train_d_' + str(dimension)
train_prefix += '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
graph_prefix += '_rec_' + str(rec_per_user) + '_v_' + str(num_votes)
interactions_file = get_file_name(train_prefix, num_users, train_partition)
print 'interaction file', interactions_file
graph_file = get_file_name(graph_prefix, num_users, train_partition, 'gml')
dataset_path = "YOUR_PATH" + dataset + "-data/"
res_path = "YOUR_PATH" + dataset + "-data/"
# load item names
items_name = {}
item_names_file = 'id_link_map.txt' # items.txt file with reversed order of columns
delimiter = '\t'
name_location = 1
with open(dataset_path + item_names_file, 'r') as f_in:
next(f_in)
for line in f_in:
tabs = line.strip().split(delimiter)