def kfold():
# ########################
# LOAD INPUT
# ########################
data_path = r'C:\Users\win10\Desktop\Projects\CYB\Experiment_Balint\CYB004\Data'
overlap = 2
freq_factor = 20
n_channels = 8
diff = "w" in [f for f in os.listdir(data_path) if f.endswith('.json')][0]
stride = 1
with open(data_path + r'\ecdfs.pickle', 'rb') as handle:
ecdfs = pickle.load(handle)
X, Y, files = data_proc(data_path, ecdf_norm, params=ecdfs, overlap=overlap, diff=diff,
window_size=freq_factor, n_channels=n_channels, task='Stair', stride=stride)
print('Data loaded. Beginning training.')
# ########################
# TRAIN NETWORK
# ########################
k = 5
drop = 0.5
kernel = (3, 3)
dil = 0
poolsize = 2
ep, ba = 100, 128
scores, model = train_net(X, Y, k=k, dil=dil, poolsize=poolsize, kernel=kernel, drop=drop, ep=ep, ba=ba)
summary(k, scores, kernel, drop, model, data_path, ep, ba, files)
def train_n_time():
# ########################
# LOAD INPUT
# ########################
data_path = r'C:\Users\win10\Desktop\Projects\CYB\Experiment_Balint\CYB004\Data'
overlap = 2
freq_factor = 20
n_channels = 8
stride = 1
diff = "w" in [f for f in os.listdir(data_path) if f.endswith('.json')][0]
with open(data_path + r'\ecdfs.pickle', 'rb') as handle:
ecdfs = pickle.load(handle)
X, Y, files = data_proc(data_path, ecdf_norm, params=ecdfs, overlap=overlap, diff=diff,
window_size=freq_factor, n_channels=n_channels, task='Stair', stride=stride)
print('Data loaded. Beginning training.')
k = 5
drop = 0.5
kernel = (3, 3)
dil = 0
poolsize = 2
ep, ba = 100, 128
from Archive.convmemnet import conv1d_model
from functools import partial
import json
from Archive.convmemnet import stack_emg
cur_model = partial(conv1d_model, n_timesteps=X.shape[1], n_features=X.shape[2],
n_outputs=Y.shape[1], drp=drop, krnl=kernel, dilate=dil, mpool=poolsize)
model = cur_model()
joint_names = ['LHip', 'RHip', 'LKnee', 'RKnee', 'LAnkle', 'RAnkle']
X0 = np.empty((0, overlap * freq_factor, n_channels))
Y0 = [[] for _ in range(len(joint_names))]
for file in sorted([f for f in os.listdir(data_path) if f.endswith('.json')]):
if 'Stair36' not in file:
continue
files.append(file)
with open(data_path + '\\' + file) as json_file:
dict_data = json.load(json_file)
for i, joint in enumerate(joint_names):
cur_data = dict_data[joint]
Y0[i].extend(cur_data[(overlap - 1):])
emg_data = ecdf_norm(np.array(dict_data["EMG"]), ecdfs)
emg_data = stack_emg(emg_data, freq_factor, overlap)
X0 = np.vstack((X0, emg_data))
Y0 = np.array(Y0)
Y0 = Y0[:, :, 0].transpose()
model.fit(X, Y, batch_size=ba, epochs=ep, verbose=2, callbacks=None, validation_data=(X0, Y0))
import time
start = time.time()
for _ in range(10):
Y1 = model.predict(X0)
end = time.time()
print(end - start)
import matplotlib.pyplot as plt
plt.plot(Y1)
plt.show()
def kfold():
# ########################
# LOAD INPUT
# ########################
data_path = r'C:\Users\win10\Desktop\Projects\CYB\Experiment_Balint\CYB004\Data'
overlap = 2
freq_factor = 20
n_channels = 8
diff = "w" in [f for f in os.listdir(data_path) if f.endswith('.json')][0]
with open(data_path + r'\lambdas.csv') as f:
reader = csv.reader(f)
params = list(reader)
params = [[float(el) for el in row] for row in params]
X, Y, files = data_proc(data_path,
box_cox_norm,
params=params,
overlap=overlap,
diff=diff,
window_size=freq_factor,
n_channels=n_channels,
task='Stair')
print('Data loaded. Beginning training.')
# ########################
# TRAIN NETWORK
# ########################
k = 5
drop = 0.5
kernel = (3, 3)
dil = 3
poolsize = 2
ep, ba = 100, 64
scores, model = train_net(X,
Y,
k=k,
dil=dil,
poolsize=poolsize,
kernel=kernel,
drop=drop,
ep=ep,
ba=ba)
summary(k, scores, kernel, drop, model, data_path, ep, ba, files)
def kfold():
# ########################
# LOAD INPUT
# ########################
data_path = r'C:\Users\hbkm9\Documents\Projects\CYB\Balint\CYB104\Data'
window_size = 240
n_channels = 8
stride = 1
freq_factor = 20
windows = (100, 100)
diff = "w" in [f for f in os.listdir(data_path) if f.endswith('.json')][0]
X, Y, files = data_proc(data_path,
norm_emg,
diff=diff,
window_size=window_size,
n_channels=n_channels,
task='Walk',
stride=stride,
windows=windows)
# X = X[:, :, :-1]
# Y = np.expand_dims(Y[:, 2], 1)
Y = normalize(np.array(Y), axis=1)
for i in range(2):
X[i] = np.expand_dims(X[i], 1)
print('Data loaded. Beginning training.')
# ########################
# TRAIN NETWORK
# ########################
k = 5
drop = 0.5
kernel = (15, 3)
dil = 3
poolsize = 4
ep, ba = 25, 20
validate = False
if validate:
scores, model = train_net(X,
Y,
k=k,
dil=dil,
poolsize=poolsize,
kernel=kernel,
drop=drop,
ep=ep,
ba=ba,
validate=validate,
window_size=window_size,
stride=stride,
freq_factor=freq_factor,
windows=windows)
summary(k, scores, kernel, drop, model, data_path, ep, ba, files)
else:
train_net(X,
Y,
k=k,
dil=dil,
poolsize=poolsize,
kernel=kernel,
drop=drop,
ep=ep,
ba=ba,
validate=validate,
window_size=window_size,
stride=stride,
freq_factor=freq_factor,
windows=windows)