def prepVect(min_df=2, max_features=50000, n_captions=5, n_sbu=None,
multilabel=False):
print "prepping the Word Tokenizer..."
_0, _1, trY, _3 = coco(mode='full', n_captions=n_captions)
if n_sbu:
_4, sbuY, _5 = sbuXYFilenames(n_sbu)
trY.extend(sbuY)
vect = Tokenizer(min_df=min_df, max_features=max_features)
captions = sampleCaptions(trY, n_captions)
vect.fit(captions)
if multilabel:
mlb = MultiLabelBinarizer()
mlb.fit(vect.transform(captions))
return vect, mlb
# if not multilabel:
return vect
def prepVect(min_df=2,
max_features=50000,
n_captions=5,
n_sbu=None,
multilabel=False):
print "prepping the Word Tokenizer..."
_0, _1, trY, _3 = coco(mode='full', n_captions=n_captions)
if n_sbu:
_4, sbuY, _5 = sbuXYFilenames(n_sbu)
trY.extend(sbuY)
vect = Tokenizer(min_df=min_df, max_features=max_features)
captions = sampleCaptions(trY, n_captions)
vect.fit(captions)
if multilabel:
mlb = MultiLabelBinarizer()
mlb.fit(vect.transform(captions))
return vect, mlb
# if not multilabel:
return vect
def traindecoder(
sources = ("image_vects", "word_vects")
, sources_k = ("image_vects_k", "word_vects_k")
, batch_size=128
, embedding_dim=300
, n_captions=5
):
# data should not be shuffled, as there is semantics in their placement
trX, teX, trY, teY = coco(mode="dev", batch_size=batch_size, n_captions=n_captions)
# # # # # # # # # # #
# Modeling Building #
# # # # # # # # # # #
stream = DataETL.getFinalStream(trX, trY, sources=sources, sources_k=sources_k,
batch_size=batch_size)
batch = stream.get_epoch_iterator().next()
f_emb = ModelIO.load('/home/luke/datasets/coco/predict/fullencoder_maxfeatures.50000')
import ipdb
ipdb.set_trace()
sys.path.insert(0, '../../python')
import planner as pln
import hardware as hw
import dataset
import models
import torch.nn
import torch
import time
simd_cfg_path = '../../hwcfg/simd.json'
hw_spec = hw.HardwareSpec(simd_cfg_path)
data = dataset.coco()
yolov2 = models.yolov2()
pnn = pln.Planner()
start_time = time.time()
conv_5_5 = None
conv_6_7 = None
conv_7 = None
for name, module in yolov2.named_modules():
if isinstance(module, torch.nn.Sequential):
continue
if name == 'conv_7':
data = conv_5_5
elif name == 'conv_8':
def trainend2end(
sources = ("image_vects", "word_tokens")
, batch_size=128
, embedding_dim=300
, n_captions=5
, mode='sample'
, n_sbu=None
, recurrent_unit='lstm'
):
"""Train a full end to end system, w/out the encoder/decoder model.
Like the google paper, "Show and Tell: Image Caption Generation"
Like how we did it with MNIST
"""
# data should not be shuffled, as there is semantics in their placement
trX, teX, trY, teY = coco(mode="full", batch_size=batch_size, n_captions=n_captions)
# add SBU
if n_sbu:
sbuX, sbuY, _ = sbuXYFilenames(n_sbu)
trX.extend(sbuX)
trY.extend(sbuY)
image_vects = T.matrix(sources[0])
word_tokens = T.lmatrix(sources[1])
image_vects.tag.test_value = np.zeros((2, 4096), dtype='float32')
word_tokens.tag.test_value = np.zeros((2, 15), dtype='int64')
from modelbuilding import ShowAndTell
show_and_tell = ShowAndTell(
image_dim=4096
, dim=embedding_dim
, dictionary_size=vect.n_features
, max_sequence_length=30
# , lookup_file='glove_lookup_53454.npy' # gloveglove
, recurrent_unit=recurrent_unit
, norm=True
, biases_init=Constant(0.)
, weights_init=IsotropicGaussian(0.02)
)
show_and_tell.initialize()
cost = show_and_tell.cost(image_vects, word_tokens)
cost.name = "seq_log_likelihood"
cg = ComputationGraph(cost)
name = "sbu+coco_NIC_%s_dim.%s" % (show_and_tell.recurrent_unit, embedding_dim)
savename = '/home/luke/datasets/coco/predict/%s' % name
def save_f_gen(self):
generated = show_and_tell.generate(image_vects)
f_gen = ComputationGraph(generated).get_theano_function()
ModelIO.save(f_gen, savename)
print "Generation function saved while training"
model = Model(cost)
algorithm = GradientDescent(
cost=cost
, parameters=cg.parameters
, step_rule=Adam(learning_rate=0.0002)
)
main_loop = MainLoop(
model=model
, data_stream=DataETL.getTokenizedStream(trX, trY, sources=sources,
batch_size=batch_size)
, algorithm=algorithm
, extensions=[
DataStreamMonitoring(
[cost]
, DataETL.getTokenizedStream(trX, trY, sources=sources,
batch_size=batch_size)
, prefix='train')
, DataStreamMonitoring(
[cost]
, DataETL.getTokenizedStream(teX, teY, sources=sources,
batch_size=batch_size)
, prefix='test')
, Printing()
, UserFunc(save_f_gen, after_epoch=True)
, FinishIfNoImprovementAfter(notification_name="test_seq_log_likelihood",
iterations=1000)
]
)
main_loop.run()
# Training finished; save the generator function w/ learned params
generated = show_and_tell.generate(image_vects)
# Beam Search
if mode == "beam":
samples, = VariableFilter(
applications=[show_and_tell.generator.generate], name="outputs")(
ComputationGraph(generated)) # generated[1] is next_outputs
beam_search = BeamSearch(samples)
try:
path = '/home/luke/datasets/coco/predict/'
filename = 'end2end_beam_maxseqlen.30_embeddingdim.300'
ModelIO.save(beam_search, '%s%s' % (path, filename))
print "It saved! Thanks pickle!"
except Exception, e:
print "F**k pickle and move on with your life :)"
print e
ModelEval.beamsearch(beam_search)
def trainencoder(
sources = ("image_vects", "word_vects")
, sources_k = ("image_vects_k", "word_vects_k")
, batch_size=128
, embedding_dim=300
, n_captions=5
, n_sbu=None
):
# data should not be shuffled, as there is semantics in their placement
trX, teX, trY, teY = coco(mode='full', batch_size=batch_size, n_captions=n_captions)
# add SBU
if n_sbu:
sbuX, sbuY, _ = sbuXYFilenames(n_sbu)
trX.extend(sbuX)
trY.extend(sbuY)
# # # # # # # # # # #
# Modeling Building #
# # # # # # # # # # #
s = Encoder(
image_feature_dim=4096
, embedding_dim=embedding_dim
, biases_init=Constant(0.)
, weights_init=Uniform(width=0.08)
)
s.initialize()
image_vects = T.matrix(sources[0]) # named to match the source name
word_vects = T.tensor3(sources[1]) # named to match the source name
image_vects_k = T.matrix(sources_k[0]) # named to match the contrastive source name
word_vects_k = T.tensor3(sources_k[1]) # named to match the contrastive source name
# image_vects.tag.test_value = np.zeros((2, 4096), dtype='float32')
# word_vects.tag.test_value = np.zeros((2, 15, 50), dtype='float32')
# image_vects_k.tag.test_value = np.zeros((2, 4096), dtype='float32')
# word_vects_k.tag.test_value = np.zeros((2, 15, 50), dtype='float32')
# learned image embedding, learned sentence embedding
lim, ls = s.apply(image_vects, word_vects)
# learned constrastive im embedding, learned contrastive s embedding
lcim, lcs = s.apply(image_vects_k, word_vects_k)
# l2norms
lim = l2norm(lim)
lcim = l2norm(lcim)
ls = l2norm(ls)
lcs = l2norm(lcs)
margin = 0.2 # alpha term, should not be more than 1!
# pairwise ranking loss (https://github.com/youralien/skip-thoughts/blob/master/eval_rank.py)
cost_im = margin - (lim * ls).sum(axis=1) + (lim * lcs).sum(axis=1)
cost_im = cost_im * (cost_im > 0.) # this is like the max(0, pairwise-ranking-loss)
cost_im = cost_im.sum(0)
cost_s = margin - (ls * lim).sum(axis=1) + (ls * lcim).sum(axis=1)
cost_s = cost_s * (cost_s > 0.) # this is like max(0, pairwise-ranking-loss)
cost_s = cost_s.sum(0)
cost = cost_im + cost_s
cost.name = "pairwise_ranking_loss"
# function to produce embedding
f_emb = theano.function([image_vects, word_vects], [lim, ls])
if n_sbu:
sbuname = "sbu.%d" % n_sbu
else:
sbuname = ''
name = "%s+coco_encoder_lstm_dim.%s_adadelta" % (sbuname, embedding_dim)
savename = '/home/luke/datasets/coco/predict/%s' % name
def save_function(self):
ModelIO.save(f_emb, savename)
print "Similarity Embedding function saved while training"
def rank_function(self):
# Get 1000 images / captions to test rank
stream = DataETL.getFinalStream(teX, teY, sources=sources,
sources_k=sources_k, batch_size=1000,
shuffle=True)
images, captions, _0, _1 = stream.get_epoch_iterator().next()
image_embs, caption_embs = f_emb(images, captions)
ModelEval.ImageSentenceRanking(image_embs, caption_embs)
cg = ComputationGraph(cost)
# # # # # # # # # # #
# Modeling Training #
# # # # # # # # # # #
algorithm = GradientDescent(
cost=cost
, parameters=cg.parameters
# , step_rule=Adam(learning_rate=0.0002)
, step_rule=AdaDelta()
)
main_loop = MainLoop(
model=Model(cost)
, data_stream=DataETL.getFinalStream(trX, trY, sources=sources,
sources_k=sources_k, batch_size=batch_size)
, algorithm=algorithm
, extensions=[
DataStreamMonitoring(
[cost]
, DataETL.getFinalStream(trX, trY, sources=sources,
sources_k=sources_k, batch_size=batch_size)
, prefix='train')
, DataStreamMonitoring(
[cost]
, DataETL.getFinalStream(teX, teY, sources=sources,
sources_k=sources_k, batch_size=batch_size)
, prefix='test')
, UserFunc(save_function, after_epoch=True)
, UserFunc(rank_function, after_epoch=True)
, Printing()
, FinishIfNoImprovementAfter(notification_name="test_pairwise_ranking_loss",
iterations=500)
]
)
main_loop.run()
# ModelIO.save(f_emb, '/home/luke/datasets/coco/predict/fullencoder_maxfeatures.50000_epochsampler')
ModelIO.save(f_emb, savename)
def train():
"""Train SqueezeDet model"""
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
with tf.Graph().as_default():
assert FLAGS.net == 'squeezeDet' or FLAGS.net == 'squeezeDet+' or FLAGS.net == 'squeezeDetSmall', \
'Selected neural net architecture not supported: {}'.format(FLAGS.net)
if FLAGS.dataset == 'COCO':
mc = coco_config()
print("COCO")
elif FLAGS.dataset == 'KITTI':
mc = kitti_squeezeDet_config()
print("KITTI")
elif FLAGS.dataset == 'BALL':
mc = ball_config()
print("BALL")
if FLAGS.net == 'squeezeDet':
mc.IS_TRAINING = True
mc.PRETRAINED_MODEL_PATH = FLAGS.pretrained_model_path
model = SqueezeDet(mc)
elif FLAGS.net == 'squeezeDet+':
mc.IS_TRAINING = True
mc.PRETRAINED_MODEL_PATH = FLAGS.pretrained_model_path
model = SqueezeDetPlus(mc)
elif FLAGS.net == 'squeezeDetSmall':
mc.IS_TRAINING = True
mc.PRETRAINED_MODEL_PATH = FLAGS.pretrained_model_path
model = SqueezeDetSmall(mc)
if FLAGS.dataset == 'COCO':
imdb = coco(FLAGS.image_set, FLAGS.data_path, mc)
elif FLAGS.dataset == 'KITTI':
imdb = kitti(FLAGS.image_set, FLAGS.data_path, mc)
elif FLAGS.dataset == 'BALL':
imdb= ball(FLAGS.image_set, FLAGS.data_path, mc)
# save model size, flops, activations by layers
with open(os.path.join(FLAGS.train_dir, 'model_metrics.txt'), 'w') as f:
f.write('Number of parameter by layer:\n')
count = 0
for c in model.model_size_counter:
f.write('\t{}: {}\n'.format(c[0], c[1]))
count += c[1]
f.write('\ttotal: {}\n'.format(count))
count = 0
f.write('\nActivation size by layer:\n')
for c in model.activation_counter:
f.write('\t{}: {}\n'.format(c[0], c[1]))
count += c[1]
f.write('\ttotal: {}\n'.format(count))
count = 0
f.write('\nNumber of flops by layer:\n')
for c in model.flop_counter:
f.write('\t{}: {}\n'.format(c[0], c[1]))
count += c[1]
f.write('\ttotal: {}\n'.format(count))
f.close()
print ('Model statistics saved to {}.'.format(
os.path.join(FLAGS.train_dir, 'model_metrics.txt')))
def _load_data(load_to_placeholder=True):
# read batch input
image_per_batch, label_per_batch, box_delta_per_batch, aidx_per_batch, \
bbox_per_batch = imdb.read_batch()
label_indices, bbox_indices, box_delta_values, mask_indices, box_values, \
= [], [], [], [], []
aidx_set = set()
num_discarded_labels = 0
num_labels = 0
for i in range(len(label_per_batch)): # batch_size
for j in range(len(label_per_batch[i])): # number of annotations
num_labels += 1
if (i, aidx_per_batch[i][j]) not in aidx_set:
aidx_set.add((i, aidx_per_batch[i][j]))
label_indices.append(
[i, aidx_per_batch[i][j], label_per_batch[i][j]])
mask_indices.append([i, aidx_per_batch[i][j]])
bbox_indices.extend(
[[i, aidx_per_batch[i][j], k] for k in range(4)])
box_delta_values.extend(box_delta_per_batch[i][j])
box_values.extend(bbox_per_batch[i][j])
else:
num_discarded_labels += 1
if mc.DEBUG_MODE:
print ('Warning: Discarded {}/({}) labels that are assigned to the same '
'anchor'.format(num_discarded_labels, num_labels))
if load_to_placeholder:
image_input = model.ph_image_input
input_mask = model.ph_input_mask
box_delta_input = model.ph_box_delta_input
box_input = model.ph_box_input
labels = model.ph_labels
else:
image_input = model.image_input
input_mask = model.input_mask
box_delta_input = model.box_delta_input
box_input = model.box_input
labels = model.labels
feed_dict = {
image_input: image_per_batch,
input_mask: np.reshape(
sparse_to_dense(
mask_indices, [mc.BATCH_SIZE, mc.ANCHORS],
[1.0]*len(mask_indices)),
[mc.BATCH_SIZE, mc.ANCHORS, 1]),
box_delta_input: sparse_to_dense(
bbox_indices, [mc.BATCH_SIZE, mc.ANCHORS, 4],
box_delta_values),
box_input: sparse_to_dense(
bbox_indices, [mc.BATCH_SIZE, mc.ANCHORS, 4],
box_values),
labels: sparse_to_dense(
label_indices,
[mc.BATCH_SIZE, mc.ANCHORS, mc.CLASSES],
[1.0]*len(label_indices)),
}
return feed_dict, image_per_batch, label_per_batch, bbox_per_batch
def _enqueue(sess, coord):
try:
while not coord.should_stop():
feed_dict, _, _, _ = _load_data()
sess.run(model.enqueue_op, feed_dict=feed_dict)
if mc.DEBUG_MODE:
print ("added to the queue")
if mc.DEBUG_MODE:
print ("Finished enqueue")
except Exception as e:
coord.request_stop(e)
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)
init = tf.global_variables_initializer()
sess.run(init)
coord = tf.train.Coordinator()
if mc.NUM_THREAD > 0:
enq_threads = []
for _ in range(mc.NUM_THREAD):
enq_thread = threading.Thread(target=_enqueue, args=[sess, coord])
# enq_thread.isDaemon()
enq_thread.start()
enq_threads.append(enq_thread)
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
run_options = tf.RunOptions(timeout_in_ms=60000)
# try:
for step in xrange(FLAGS.max_steps):
if coord.should_stop():
sess.run(model.FIFOQueue.close(cancel_pending_enqueues=True))
coord.request_stop()
coord.join(threads)
break
start_time = time.time()
if step % FLAGS.summary_step == 0:
feed_dict, image_per_batch, label_per_batch, bbox_per_batch = \
_load_data(load_to_placeholder=False)
op_list = [
model.train_op, model.loss, summary_op, model.det_boxes,
model.det_probs, model.det_class, model.conf_loss,
model.bbox_loss, model.class_loss
]
_, loss_value, summary_str, det_boxes, det_probs, det_class, \
conf_loss, bbox_loss, class_loss = sess.run(
op_list, feed_dict=feed_dict)
_viz_prediction_result(
model, image_per_batch, bbox_per_batch, label_per_batch, det_boxes,
det_class, det_probs)
image_per_batch = bgr_to_rgb(image_per_batch)
viz_summary = sess.run(
model.viz_op, feed_dict={model.image_to_show: image_per_batch})
summary_writer.add_summary(summary_str, step)
summary_writer.add_summary(viz_summary, step)
summary_writer.flush()
print ('conf_loss: {}, bbox_loss: {}, class_loss: {}'.
format(conf_loss, bbox_loss, class_loss))
else:
if mc.NUM_THREAD > 0:
_, loss_value, conf_loss, bbox_loss, class_loss = sess.run(
[model.train_op, model.loss, model.conf_loss, model.bbox_loss,
model.class_loss], options=run_options)
else:
feed_dict, _, _, _ = _load_data(load_to_placeholder=False)
_, loss_value, conf_loss, bbox_loss, class_loss = sess.run(
[model.train_op, model.loss, model.conf_loss, model.bbox_loss,
model.class_loss], feed_dict=feed_dict)
duration = time.time() - start_time
assert not np.isnan(loss_value), \
'Model diverged. Total loss: {}, conf_loss: {}, bbox_loss: {}, ' \
'class_loss: {}'.format(loss_value, conf_loss, bbox_loss, class_loss)
if step % 10 == 0:
num_images_per_step = mc.BATCH_SIZE
images_per_sec = num_images_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f images/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), step, loss_value,
images_per_sec, sec_per_batch))
sys.stdout.flush()
# Save the model checkpoint periodically.
if step % FLAGS.checkpoint_step == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
def trainencoder(
sources = ("image_vects", "word_vects")
, sources_k = ("image_vects_k", "word_vects_k")
, batch_size=128
, embedding_dim=300
, n_captions=5
, n_sbu=None
, separate_emb=False
, test_size=1000 # per dataset
, mode='dev'
):
if mode=="coco120k+flickr38k":
XYsplit_cum = ([], [], [], [])
xyloaders = [
"cocoXYFilenames(dataType='train2014')"
, "cocoXYFilenames(dataType='val2014')"
, "flickrXYFilenames(dataType='8k')"
, "flickrXYFilenames(dataType='30k')"
]
ntrains = [80000, 40000, 8000, 30000]
for xyloader, ntrain in zip(xyloaders, ntrains):
X, Y, _ = eval(xyloader)
XYsplit = train_test_split(X, Y, train_size=ntrain)
for i in range(len(XYsplit)):
XYsplit_cum[i].extend(XYsplit[i])
trX, teX, trY, teY = XYsplit_cum
else:
trX, teX, trY, teY = coco(mode=mode, n_captions=n_captions, test_size=test_size)
if n_sbu:
sbutrX, sbuteX, sbutrY, sbuteY = sbu(mode=mode, test_size=test_size)
pairs = (
(trX, sbutrX)
, (teX, sbuteX)
, (trY, sbutrY)
, (teY, sbuteY)
)
for coco_data, sbu_data in pairs:
if isinstance(coco_data, list):
coco_data.extend(sbu_data)
print("n_train: %d" % len(trX))
print("n_test: %d" % len(teX))
# # # # # # # # # # #
# Modeling Building #
# # # # # # # # # # #
s = Encoder(
image_feature_dim=4096
, embedding_dim=embedding_dim
, biases_init=Constant(0.)
, weights_init=Uniform(width=0.08)
)
s.initialize()
image_vects = tensor.matrix(sources[0]) # named to match the source name
word_vects = tensor.tensor3(sources[1]) # named to match the source name
image_vects_k = tensor.matrix(sources_k[0]) # named to match the contrastive source name
word_vects_k = tensor.tensor3(sources_k[1]) # named to match the contrastive source name
# image_vects.tag.test_value = np.zeros((2, 4096), dtype='float32')
# word_vects.tag.test_value = np.zeros((2, 15, 50), dtype='float32')
# image_vects_k.tag.test_value = np.zeros((2, 4096), dtype='float32')
# word_vects_k.tag.test_value = np.zeros((2, 15, 50), dtype='float32')
# learned image embedding, learned sentence embedding
lim, ls = s.apply(image_vects, word_vects)
# learned constrastive im embedding, learned contrastive s embedding
lcim, lcs = s.apply(image_vects_k, word_vects_k)
# identical cost code thanks to Ryan Kiros
# https://github.com/youralien/skip-thoughts/blob/master/eval_rank.py
lim = l2norm(lim)
lcim = l2norm(lcim)
ls = l2norm(ls)
lcs = l2norm(lcs)
margin = 0.2 # alpha term should not be more than 1
cost_im = margin - (lim * ls).sum(axis=1) + (lim * lcs).sum(axis=1)
cost_im = cost_im * (cost_im > 0.) # this is like the max(0, pairwise-ranking-loss)
cost_im = cost_im.sum(0)
cost_s = margin - (ls * lim).sum(axis=1) + (ls * lcim).sum(axis=1)
cost_s = cost_s * (cost_s > 0.) # this is like max(0, pairwise-ranking-loss)
cost_s = cost_s.sum(0)
cost = cost_im + cost_s
cost.name = "pairwise_ranking_loss"
# function(s) to produce embedding
if separate_emb:
img_encoder = theano.function([image_vects], lim)
txt_encoder = theano.function([word_vects], ls)
f_emb = theano.function([image_vects, word_vects], [lim, ls])
if n_sbu:
sbuname = "sbu%d+" % n_sbu
else:
sbuname = ''
name = "%sproject1.%s.jointembedder" % (sbuname, mode)
savename = MODEL_FILES_DIR + name
def save_function(self):
if separate_emb:
ModelIO.save(
img_encoder
, savename + "_Img")
ModelIO.save(
txt_encoder
, savename + "_Txt")
ModelIO.save(f_emb, savename)
print "Similarity Embedding function(s) saved while training"
def rank_function(stream):
images, captions, _0, _1 = stream.get_epoch_iterator().next()
image_embs, caption_embs = f_emb(images, captions)
ModelEval.ImageSentenceRanking(image_embs, caption_embs)
def rank_coco(self=None):
# Get 1000 images / captions to test rank
stream = DataETL.getFinalStream(teX, teY, sources=sources,
sources_k=sources_k, batch_size=test_size,
shuffle=True)
print "COCO test"
rank_function(stream)
def rank_sbu(self=None):
stream = DataETL.getFinalStream(sbuteX, sbuteY, sources=sources,
sources_k=sources_k, batch_size=test_size,
shuffle=True)
print "SBU test"
rank_function(stream)
def rank_em(self=None):
rank_coco()
if n_sbu:
rank_sbu()
cg = ComputationGraph(cost)
# # # # # # # # # # #
# Modeling Training #
# # # # # # # # # # #
algorithm = GradientDescent(
cost=cost
, parameters=cg.parameters
, step_rule=Adam(learning_rate=0.0002)
)
main_loop = MainLoop(
model=Model(cost)
, data_stream=DataETL.getFinalStream(trX, trY, sources=sources,
sources_k=sources_k, batch_size=batch_size)
, algorithm=algorithm
, extensions=[
DataStreamMonitoring(
[cost]
, DataETL.getFinalStream(trX, trY, sources=sources,
sources_k=sources_k, batch_size=batch_size, shuffle=True)
, prefix='train')
, DataStreamMonitoring(
[cost]
, DataETL.getFinalStream(teX, teY, sources=sources,
sources_k=sources_k, batch_size=batch_size, shuffle=True)
, prefix='test')
, UserFunc(save_function, after_epoch=True)
, UserFunc(rank_em, after_epoch=True)
, Printing()
, LogToFile('logs/%s.csv' % name)
]
)
main_loop.run()