def show_frames(frame_path, dir_path):
paths = files.get_paths(dir_path, ["actions", "bounds"])
action_imgs.mean_action(frame_path, paths["actions"], dims=None)
tools.cut_actions(paths["actions"],
paths["bounds"],
scale=None,
binary=True)
def reg_pretrain(in_path, reg_epochs=3000, clf_epochs=1000):
post = ['seqs', 'spline', 'stats', 'feats', 'reg_nn', 'clf_nn']
paths = files.get_paths(in_path, post)
spline.ens_upsample(paths['seqs'], paths["spline"])
stats.ens_stats(paths['seqs'], paths['stats'])
train_reg(paths['spline'], paths['stats'], paths['reg_nn'], 3000)
pretrain_clf(paths['reg_nn'], paths['spline'], paths['clf_nn'], clf_epochs)
basic.ts.ens_extract(paths["spline"], paths["clf_nn"], paths["feats"])
def agum_exp(input_path, out_path, n_epochs=1000):
files.make_dir(out_path)
paths = files.get_paths(out_path, ["spline", "agum", "nn", "feats"])
spline.upsample(input_path, paths["spline"], size=64)
agum = Agum([ts_scale, magnitude_warp])
agum(paths["spline"], paths["agum"])
train, extract = ts_cnn.get_train()
train(paths["agum"], paths["nn"], n_epochs)
extract(paths["agum"], paths["nn"], paths["feats"])
def voting(in_path):
paths = files.get_paths(in_path, name="pairs")
dtw_pairs = read_pairs(paths)
pref_dicts = [get_preferences(pairs_i) for pairs_i in dtw_pairs]
names = pref_dicts[0].keys()
y_pred, y_true = [], []
for name_i in names:
y_true.append(name_i.get_cat())
votes = [pref_j[name_i] for pref_j in pref_dicts]
y_pred.append(border_count(votes))
print(accuracy_score(y_true, y_pred))
def __call__(self, input_paths, out_path, arg_dict):
dirs = files.get_paths(out_path, self.dir_names)
for dir_i in dirs.values():
files.make_dir(dir_i)
for path_i in input_paths:
name_i = path_i.split('/')[-1]
args_i = {
key_i: "%s/%s" % (path_i, name_i)
for key_i, path_i in dirs.items()
}
args_i = {**args_i, **arg_dict}
args_i[self.input_dir] = path_i
for fun, arg_names in self.funcs:
fun_args = [args_i[name_k] for name_k in arg_names]
fun(*fun_args)
def basic_paths(dataset, dir_path, common, binary, name="dtw"):
paths = {}
paths["dir_path"] = "%s/%s" % (dir_path, dataset)
common = "%s/%s" % (paths["dir_path"], common)
paths["common"] = files.get_paths(common, name=name)
if (binary):
paths["binary"] = "%s/%s" % (paths["dir_path"], binary)
else:
paths["binary"] = None
return paths
#def common_paths(common,binary):
# common=files.top_files(common)
# datasets=[ common_i.split('/')[-1].split("_")[0]
# for common_i in common]
# binary=[ binary % data_i for data_i in datasets]
# return list(zip(common,binary))
def ae_exp(in_path, out_path, n_epochs=5):
files.make_dir(out_path)
paths = files.get_paths(out_path, ["ae", "seqs"])
train_ae(in_path, paths['ae'], n_epochs)
extract(in_path, paths['ae'], paths['seqs'])
default=config.NUM_SCREENSHOTS,
help='number of screenshots to save and upload',
)
parser.add_argument(
'-d',
'--delete-unwanted-files',
dest='delete_unwanted_files',
action='store_true',
default=config.DELETE_UNWANTED_FILES,
help='delete files that are not whitelisted by the tracker (such as .rar files)',
)
args = parser.parse_args()
# Every argument to this script is treated as a path to a release to be uploaded
release_list = files.get_paths(args.file_list)
if len(release_list) == 0:
logging.critical('You must give this script at least one file or directory to process!')
sys.exit(1)
for path in release_list:
# Log exceptions but don't raise them; just continue
try:
if isinstance(path, bytes):
path = path.decode('utf-8')
files.set_log_file_name(os.path.basename(path) + '.log')
upload = uploads.Upload(
def action_img_exp(in_path, n_epochs=100):
paths = files.get_paths(in_path, ['frames', 'mean', "ens"])
mean_action(paths["frames"], paths["mean"])
action_one_shot(paths["mean"], paths["action"], n_epochs)
def make_reduced_dataset(common, out_path, n_feats=350):
common = files.get_paths(common)
dataset = ens.read_dataset(common, None)[0]
dataset.norm()
redu_data = reduction.reduce(dataset, n_feats)
redu_data.save(out_path)
def lstm_exp(in_path, out_path, n_epochs=200, seq_len=20, static=True):
files.make_dir(out_path)
paths = files.get_paths(out_path, ["nn", "feats"])
train_lstm(in_path, paths['nn'], n_epochs, seq_len, static=static)
extract(in_path, paths['nn'], paths['feats'], seq_len)
def floor_exp(action_path, frame_path, dataset_path, out_path, n_epochs=750):
files.make_dir(out_path)
paths = files.get_paths(out_path, ["nn", "frames", "actions"])
train(action_path, dataset_path, paths["nn"], n_epochs)
apply_model(frame_path, paths["nn"], paths["frames"])
action_imgs.mean_action(paths["frames"], paths["actions"], None)
def agum_exp(in_path, n_epochs=100):
paths = files.get_paths(in_path, ['box', 'filtr_nn', 'frames'])
train_model(paths["box"], paths["filtr_nn"], n_epochs)
filtr_seqs(paths["box"], paths["filtr_nn"], paths["frames"])
default=config.NUM_SCREENSHOTS,
help='number of screenshots to save and upload',
)
parser.add_argument(
'-d',
'--delete-unwanted-files',
dest='delete_unwanted_files',
action='store_true',
default=config.DELETE_UNWANTED_FILES,
help=
'delete files that are not whitelisted by the tracker (such as .rar files)',
)
args = parser.parse_args()
# Every argument to this script is treated as a path to a release to be uploaded
release_list = files.get_paths(args.file_list)
if len(release_list) == 0:
logging.critical(
'You must give this script at least one file or directory to process!')
sys.exit(1)
for path in release_list:
# Log exceptions but don't raise them; just continue
try:
if isinstance(path, bytes):
path = path.decode('utf-8')
files.set_log_file_name(os.path.basename(path) + '.log')
clf = linear_model.Lasso(alpha=0.001,max_iter=1000)
clf.fit(X_train,y_train)
model = SelectFromModel(clf, prefit=True)
new_X= model.transform(X)
return new_X
def recursive(train_i,full_i,n=84):
X_train,y_train,names_train=train_i.as_dataset()
svc = SVC(kernel='linear',C=1)
rfe = RFE(estimator=svc,n_features_to_select=n,step=10)
rfe.fit(X_train,y_train)
X,y,names=full_i.as_dataset()
new_X= rfe.transform(X)
return new_X
def selected_common(common_path,out_path,n=100):
dataset=ens.read_dataset(common_path,None)[0]
dataset.norm()
new_data=reduce(dataset,n)
new_data.save(out_path)
if __name__ == "__main__":
dataset="MHAD"
dir_path="../ICSS_exp/%s" % dataset
common="%s/dtw" % dir_path
common=files.get_paths(common,name="dtw")
common.append("%s/common/1D_CNN/feats" % dir_path)
binary="%s/ens/lstm/feats" % dir_path
acc=acc_curve(common,binary,clf="SVC",n=20,step=50)
n=150
# selected_common(common,"s_feats/%s_%d" % (dataset,n),n)
def sampled_exp(in_path, out_path, n_samples=3, n_epochs=5):
files.make_dir(out_path)
paths = files.get_paths(out_path, ["nn", "seqs"])
train(in_path, paths["nn"], n_samples, n_epochs)
extract(in_path, paths['nn'], paths['seqs'])
def basic_exp(in_path):
paths = files.get_paths(in_path, name="dtw")
result = ens.ensemble(paths[0], None, clf="LR", binary=False)[0]
result.report()
def dtw_exp(seq_path,out_path): files.make_dir(out_path) paths=files.get_paths(out_path,["pairs","dtw"]) dtw.pairs.compute_pairs(seq_path,paths["pairs"]) compute_feats(paths["pairs"],paths["dtw"])