def evaluate(data):
sum_err = 0.0
for it in xrange(0, len(data), batch_size):
batch = data[it:it + batch_size]
if len(batch) < batch_size:
batch.extend(data[0:batch_size - len(batch)])
src_batch, label_batch = zip(*batch)
src_tuple = smiles2graph_list(src_batch)
feed_map = {x: y for x, y in zip(src_holder, src_tuple)}
feed_map.update({label: label_batch})
err = session.run(loss, feed_dict=feed_map)
sum_err += err
return math.sqrt(sum_err / len(data))
buckets.pop(i)
sum_err, sum_gnorm = 0.0, 0.0
it = 0
_lr = 0.001
while True:
it += batch_size
batch = []
for i in xrange(batch_size):
buk = random.choice(buckets)
idx = buk.popleft()
batch.append(train[idx])
buk.append(idx)
src_batch, label_batch = zip(*batch)
src_tuple = smiles2graph_list(src_batch)
feed_map = {x: y for x, y in zip(src_holder, src_tuple)}
feed_map.update({label: label_batch, lr: _lr})
_, err, pnorm, gnorm = session.run([backprop, loss, param_norm, grad_norm],
feed_dict=feed_map)
sum_err += err
sum_gnorm += gnorm
if it % 200 == 0 and it > 0:
rmse = math.sqrt(sum_err / 200)
print "Training RMSE: %.4f, Param Norm: %.2f, Grad Norm: %.2f" % (
rmse, pnorm, sum_gnorm / 200)
sys.stdout.flush()
sum_err, sum_gnorm = 0.0, 0.0
if it % 100000 == 0 and it > 0:
