def dump(dataset, options):
"""
:type dataset: Dataset.Dataset
:param options: argparse.Namespace
"""
print("Epoch: %i" % options.epoch, file=log.v3)
rnn.train_data.init_seq_order(options.epoch)
output_dict = {}
for name, layer in rnn.engine.network.layers.items():
output_dict["%s:out" % name] = layer.output.placeholder
for i, v in layer.output.size_placeholder.items():
output_dict["%s:shape(%i)" %
(name, layer.output.get_batch_axis(i))] = v
seq_idx = options.startseq
if options.endseq < 0:
options.endseq = float("inf")
while dataset.is_less_than_num_seqs(seq_idx) and seq_idx <= options.endseq:
print("Seq idx: %i" % (seq_idx, ), file=log.v3)
out = rnn.engine.eval_single(dataset=dataset,
seq_idx=seq_idx,
output_dict=output_dict)
for name, v in sorted(out.items()):
print(" %s: %s" % (name, pretty_print(v)))
seq_idx += 1
print("Done. More seqs which we did not dumped: %s" %
dataset.is_less_than_num_seqs(seq_idx),
file=log.v1)
def dump(dataset, options):
"""
:type dataset: Dataset.Dataset
:param options: argparse.Namespace
"""
print("Epoch: %i" % options.epoch, file=log.v3)
dataset.init_seq_order(options.epoch)
output_dict = {}
for name, layer in rnn.engine.network.layers.items():
output_dict["%s:out" % name] = layer.output.placeholder
for i, v in layer.output.size_placeholder.items():
output_dict["%s:shape(%i)" % (name, layer.output.get_batch_axis(i))] = v
seq_idx = options.startseq
if options.endseq < 0:
options.endseq = float("inf")
while dataset.is_less_than_num_seqs(seq_idx) and seq_idx <= options.endseq:
print("Seq idx: %i" % (seq_idx,), file=log.v3)
out = rnn.engine.run_single(dataset=dataset, seq_idx=seq_idx, output_dict=output_dict)
for name, v in sorted(out.items()):
print(" %s: %s" % (name, pretty_print(v)))
seq_idx += 1
print("Done. More seqs which we did not dumped: %s" % dataset.is_less_than_num_seqs(seq_idx), file=log.v1)
def dump_dataset(dataset, options):
"""
:type dataset: Dataset.Dataset
:param options: argparse.Namespace
"""
print("Epoch: %i" % options.epoch, file=log.v3)
rnn.train_data.init_seq_order(options.epoch)
if options.type == "numpy":
print("Dump files: %r*%r" %
(options.dump_prefix, options.dump_postfix),
file=log.v3)
elif options.type == "stdout":
print("Dump to stdout", file=log.v3)
else:
raise Exception("unknown dump option type %r" % options.type)
seq_idx = options.startseq
if options.endseq < 0:
options.endseq = float("inf")
while dataset.is_less_than_num_seqs(seq_idx) and seq_idx <= options.endseq:
dataset.load_seqs(seq_idx, seq_idx + 1)
data = dataset.get_data(seq_idx, "data")
if options.type == "numpy":
numpy.savetxt(
"%s%i.data%s" %
(options.dump_prefix, seq_idx, options.dump_postfix), data)
elif options.type == "stdout":
print("seq %i data:" % seq_idx, pretty_print(data))
for target in dataset.get_target_list():
targets = dataset.get_targets(target, seq_idx)
if options.type == "numpy":
numpy.savetxt("%s%i.targets.%s%s" %
(options.dump_prefix, seq_idx, target,
options.dump_postfix),
targets,
fmt='%i')
elif options.type == "stdout":
print("seq %i target %r:" % (seq_idx, target),
pretty_print(targets))
seq_idx += 1
print("Done. More seqs which we did not dumped: %s" %
dataset.is_less_than_num_seqs(seq_idx),
file=log.v1)
def dump_dataset(dataset, options):
"""
:type dataset: Dataset.Dataset
:param options: argparse.Namespace
"""
print("Epoch: %i" % options.epoch, file=log.v3)
dataset.init_seq_order(epoch=options.epoch)
if options.get_num_seqs:
print("Get num seqs.")
print("estimated_num_seqs: %r" % dataset.estimated_num_seqs)
try:
print("num_seqs: %r" % dataset.num_seqs)
except Exception as exc:
print("num_seqs exception %r, which is valid, so we count." % exc)
seq_idx = 0
if dataset.get_target_list():
default_target = dataset.get_target_list()[0]
else:
default_target = None
while dataset.is_less_than_num_seqs(seq_idx):
dataset.load_seqs(seq_idx, seq_idx + 1)
if seq_idx % 10000 == 0:
if default_target:
targets = dataset.get_targets(default_target, seq_idx)
postfix = " (targets = %r...)" % (targets[:10], )
else:
postfix = ""
print("%i ...%s" % (seq_idx, postfix))
seq_idx += 1
print("accumulated num seqs: %i" % seq_idx)
print("Done.")
return
if options.type == "numpy":
print("Dump files: %r*%r" %
(options.dump_prefix, options.dump_postfix),
file=log.v3)
elif options.type == "stdout":
print("Dump to stdout", file=log.v3)
else:
raise Exception("unknown dump option type %r" % options.type)
seq_idx = options.startseq
if options.endseq < 0:
options.endseq = float("inf")
while dataset.is_less_than_num_seqs(seq_idx) and seq_idx <= options.endseq:
dataset.load_seqs(seq_idx, seq_idx + 1)
data = dataset.get_data(seq_idx, "data")
if options.type == "numpy":
numpy.savetxt(
"%s%i.data%s" %
(options.dump_prefix, seq_idx, options.dump_postfix), data)
elif options.type == "stdout":
print("seq %i data:" % seq_idx, pretty_print(data))
for target in dataset.get_target_list():
targets = dataset.get_targets(target, seq_idx)
if options.type == "numpy":
numpy.savetxt("%s%i.targets.%s%s" %
(options.dump_prefix, seq_idx, target,
options.dump_postfix),
targets,
fmt='%i')
elif options.type == "stdout":
print("seq %i target %r:" % (seq_idx, target),
pretty_print(targets))
seq_idx += 1
print("Done. More seqs which we did not dumped: %s" %
dataset.is_less_than_num_seqs(seq_idx),
file=log.v1)
def dump_dataset(dataset, options):
"""
:type dataset: Dataset.Dataset
:param options: argparse.Namespace
"""
print("Epoch: %i" % options.epoch, file=log.v3)
dataset.init_seq_order(epoch=options.epoch)
print("Dataset keys:", dataset.get_data_keys(), file=log.v3)
print("Dataset target keys:", dataset.get_target_list(), file=log.v3)
assert options.key in dataset.get_data_keys()
if options.get_num_seqs:
print("Get num seqs.")
print("estimated_num_seqs: %r" % dataset.estimated_num_seqs)
try:
print("num_seqs: %r" % dataset.num_seqs)
except Exception as exc:
print("num_seqs exception %r, which is valid, so we count." % exc)
seq_idx = 0
if dataset.get_target_list():
default_target = dataset.get_target_list()[0]
else:
default_target = None
while dataset.is_less_than_num_seqs(seq_idx):
dataset.load_seqs(seq_idx, seq_idx + 1)
if seq_idx % 10000 == 0:
if default_target:
targets = dataset.get_targets(default_target, seq_idx)
postfix = " (targets = %r...)" % (targets[:10], )
else:
postfix = ""
print("%i ...%s" % (seq_idx, postfix))
seq_idx += 1
print("accumulated num seqs: %i" % seq_idx)
print("Done.")
return
if options.type == "numpy":
print("Dump files: %r*%r" %
(options.dump_prefix, options.dump_postfix),
file=log.v3)
elif options.type == "stdout":
print("Dump to stdout", file=log.v3)
elif options.type == "print_shape":
print("Dump shape to stdout", file=log.v3)
elif options.type == "plot":
print("Plot.", file=log.v3)
elif options.type == "null":
print("No dump.")
else:
raise Exception("unknown dump option type %r" % options.type)
start_time = time.time()
stats = Stats() if (options.stats or options.dump_stats) else None
seq_len_stats = {key: Stats() for key in dataset.get_data_keys()}
seq_idx = options.startseq
if options.endseq < 0:
options.endseq = float("inf")
while dataset.is_less_than_num_seqs(seq_idx) and seq_idx <= options.endseq:
dataset.load_seqs(seq_idx, seq_idx + 1)
complete_frac = dataset.get_complete_frac(seq_idx)
start_elapsed = time.time() - start_time
try:
num_seqs_s = str(dataset.num_seqs)
except NotImplementedError:
try:
num_seqs_s = "~%i" % dataset.estimated_num_seqs
except TypeError: # a number is required, not NoneType
num_seqs_s = "?"
progress_prefix = "%i/%s" % (seq_idx, num_seqs_s)
progress = "%s (%.02f%%)" % (progress_prefix, complete_frac * 100)
if complete_frac > 0:
total_time_estimated = start_elapsed / complete_frac
remaining_estimated = total_time_estimated - start_elapsed
progress += " (%s)" % hms(remaining_estimated)
data = dataset.get_data(seq_idx, options.key)
if options.type == "numpy":
numpy.savetxt(
"%s%i.data%s" %
(options.dump_prefix, seq_idx, options.dump_postfix), data)
elif options.type == "stdout":
print("seq %s data:" % progress, pretty_print(data))
elif options.type == "print_shape":
print("seq %s data shape:" % progress, data.shape)
elif options.type == "plot":
plot(data)
for target in dataset.get_target_list():
targets = dataset.get_targets(target, seq_idx)
if options.type == "numpy":
numpy.savetxt("%s%i.targets.%s%s" %
(options.dump_prefix, seq_idx, target,
options.dump_postfix),
targets,
fmt='%i')
elif options.type == "stdout":
print("seq %i target %r:" % (seq_idx, target),
pretty_print(targets))
elif options.type == "print_shape":
print("seq %i target %r shape:" % (seq_idx, target),
targets.shape)
seq_len = dataset.get_seq_length(seq_idx)
for key in dataset.get_data_keys():
seq_len_stats[key].collect([seq_len[key]])
if stats:
stats.collect(data)
if options.type == "null":
Util.progress_bar_with_time(complete_frac, prefix=progress_prefix)
seq_idx += 1
print("Done. Total time %s. More seqs which we did not dumped: %s" %
(hms(time.time() - start_time),
dataset.is_less_than_num_seqs(seq_idx)),
file=log.v1)
for key in dataset.get_data_keys():
seq_len_stats[key].dump(stream_prefix="Seq-length %r " % key,
stream=log.v2)
if stats:
stats.dump(output_file_prefix=options.dump_stats,
stream_prefix="Data %r " % options.key,
stream=log.v2)