def recommend():
with tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.device_log)) as sess:
print("reading data")
task = Task(FLAGS.dataset)
(_, items_dev, _, _, u_attributes, i_attributes, item_ind2logit_ind,
logit_ind2item_ind, _) = read_data(task)
Evaluate = task.get_evaluation()
evaluation = Evaluate(logit_ind2item_ind, ta=FLAGS.ta, test=FLAGS.test)
model = create_model(sess,
u_attributes,
i_attributes,
item_ind2logit_ind,
logit_ind2item_ind,
loss=FLAGS.loss,
ind_item=None)
Uinds = evaluation.get_uinds()
N = len(Uinds)
print("N = %d" % N)
Uinds = [p for p in Uinds if p in items_dev]
mylog("new N = {}, (reduced from original {})".format(len(Uinds), N))
if len(Uinds) < N:
evaluation.set_uinds(Uinds)
N = len(Uinds)
rec = np.zeros((N, FLAGS.top_N_items), dtype=int)
count = 0
time_start = time.time()
for idx_s in range(0, N, FLAGS.batch_size):
count += 1
if count % 100 == 0:
print("idx: %d, c: %d" % (idx_s, count))
idx_e = idx_s + FLAGS.batch_size
if idx_e <= N:
users = Uinds[idx_s:idx_e]
items_input = [items_dev[u] for u in users]
items_input = map(list, zip(*items_input))
recs = model.step(sess,
users,
items_input,
forward_only=True,
recommend=True,
recommend_new=FLAGS.recommend_new)
rec[idx_s:idx_e, :] = recs
else:
users = range(idx_s, N) + [0] * (idx_e - N)
users = [Uinds[t] for t in users]
items_input = [items_dev[u] for u in users]
items_input = map(list, zip(*items_input))
recs = model.step(sess,
users,
items_input,
forward_only=True,
recommend=True,
recommend_new=FLAGS.recommend_new)
idx_e = N
rec[idx_s:idx_e, :] = recs[:(idx_e - idx_s), :]
# return rec: i: uinds[i] --> logid
time_end = time.time()
mylog("Time used %.1f" % (time_end - time_start))
R = evaluation.gen_rec(rec, FLAGS.recommend_new)
evaluation.eval_on(R)
s1, s2, s3, s4, s5 = evaluation.get_scores()
mylog('scores: \n\t{}\n\t{}\n\t{}\n'.format(s1, s2, s3))
mylog("SCORE_FORMAT: {} {} {}".format(s1[0], s2[0], s3[0]))
mylog("METRIC_FORMAT1: {}".format(s4))
mylog("METRIC_FORMAT2: {}".format(s5))
return
