def main():
parameters = arguments()
model_commands = [
## 2
"--model_type classifier --architecture rnn --data_type buckeye --epochs 50 "
+
"--features_type mfcc --train_tag gt --enc 400_400_400 --latent 130 --batch_size 512",
"--model_type classifier --architecture cnn --data_type omniglot --epochs 50 "
+
"--enc 3.3.32.1.1.2.2_3.3.64.1.1.2.2_3.3.128.1.1 --latent 130 --batch_size 64",
]
test_seeds = parameters.test_seeds == "True"
for this_command in model_commands:
if test_seeds:
for rnd_seed in SEEDS:
cmd = "./train_model.py " + this_command + " --rnd_seed {}".format(
rnd_seed)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
else:
cmd = "./train_model.py " + this_command
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
def main():
feat_fns = []
for dataset_type in SPEECH_DATASET_TYPE:
for dataset in SPEECH_DATASETS:
feat_fns.append(
path.join(feats_path, dataset, "Subsets", "Words", "mfcc",
"gt_" + dataset_type + "_mfcc.npz"))
for dataset_type in IMAGE_DATASET_TYPE:
for dataset in IMAGE_DATASETS:
feat_fns.append(
path.join(feats_path, dataset, dataset_type + ".npz"))
for (fn, num, metr) in list(itertools.product(feat_fns, NUM_PAIRS,
METRIC)):
cmd = "./classifier_latents.py " + " --feats_fn {}".format(fn)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
cmd = "./classifier_pairs.py " + " --feats_fn {} --num_pairs {} --metric {}".format(
fn, num, metr)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
def main():
parameters = arguments()
model_commands = [
## CAE hard pairs
"--model_type cae --architecture rnn --data_type TIDigits --epochs 50 "
+
"--features_type mfcc --train_tag gt --enc 400_400_400 --latent 130 --pair_type classifier",
"--model_type cae --architecture cnn --data_type MNIST --epochs 50 " +
"--enc 3.3.32.1.1.2.2_3.3.64.1.1.2.2_3.3.128.1.1 --latent 130 --pair_type classifier",
## 2
"--model_type classifier --architecture rnn --data_type buckeye --epochs 50 --one_shot_speech_dataset buckeye "
+ "--features_type mfcc --train_tag gt --enc 400_400_400 --latent 130",
"--model_type classifier --architecture cnn --data_type omniglot --epochs 50 --one_shot_image_dataset omniglot "
+ "--enc 3.3.32.1.1.2.2_3.3.64.1.1.2.2_3.3.128.1.1 --latent 130",
]
for this_command in model_commands:
for batch_size in BATCH_SIZES:
for rnd_seed in SEEDS:
cmd = "./train_model.py " + this_command + " --rnd_seed {} --batch_size {} --final_model {}".format(
rnd_seed, batch_size, False)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
def main():
feat_fns = []
for dataset_type in DATASET_TYPES:
for (speech_dataset, image_dataset) in PAIR_DATASETS:
speech_dataset_type = "val" if dataset_type == "validation" else dataset_type
speech_fn = path.join(feats_path, speech_dataset, "Subsets",
"Words", "mfcc",
"gt_" + speech_dataset_type + "_mfcc.npz")
image_fn = path.join(feats_path, image_dataset,
dataset_type + ".npz")
feat_fns.append((speech_fn, image_fn))
for ((sp_fn, im_fn), metr) in list(itertools.product(feat_fns, METRIC)):
cmd = "./classifier_latents.py " + " --feats_fn {}".format(sp_fn)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
cmd = "./classifier_latents.py " + " --feats_fn {}".format(im_fn)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
cmd = "./speech_image_pairs.py " + " --speech_feats_fn {} --image_feats_fn {} --metric {}".format(
sp_fn, im_fn, metr)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
def main():
directories = os.walk("../Model_data/")
valid_dirs = []
for root, dirs, files in directories:
for filename in files:
if filename.split("_")[-1] == "log.txt" and len(dirs) != 0:
log = path.join(root, filename)
name = root.split("/")[-1]
lib = path.join(root, dirs[0], name + "_lib.pkl")
valid_dirs.append((lib, log))
pair_logs = []
appended_logs = []
count = 0
for lib1, log1 in valid_dirs:
first_lib = model_setup_library.restore_lib(lib1)
for lib2, log2 in valid_dirs:
if lib1 != lib2 and log1 not in appended_logs and log2 not in appended_logs:
second_lib = model_setup_library.restore_lib(lib2)
if first_lib["training_on"] == "images":
image_lib = first_lib.copy()
image_log = log1
else:
speech_lib = first_lib.copy()
speech_log = log1
if second_lib["training_on"] == "images" and first_lib[
"training_on"] == "speech":
image_lib = second_lib.copy()
image_log = log2
elif second_lib["training_on"] == "speech" and first_lib[
"training_on"] == "images":
speech_lib = second_lib.copy()
speech_log = log2
else:
break
pairs = pair_check(speech_lib, image_lib)
if pairs:
pair_logs.append((speech_log, image_log))
appended_logs.append(speech_log)
appended_logs.append(image_log)
for speech_fn, image_fn in pair_logs:
cmd = "./few_shot_learning.py --speech_log_fn {} --image_log_fn {} --speech_data_fn {} --image_data_fn {} --episode_fn {}".format(
speech_fn, image_fn, speech_dataset, image_dataset,
one_shot_episode)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()
cmd = "./few_shot_learning.py --speech_log_fn {} --image_log_fn {} --speech_data_fn {} --image_data_fn {} --episode_fn {}".format(
speech_fn, image_fn, speech_dataset, image_dataset,
few_shot_episode)
print_string = cmd
model_setup_library.command_printing(print_string)
sys.stdout.flush()
proc = subprocess.Popen(cmd, shell=True)
proc.wait()