def main():
print("start running basic neural network")
np.random.seed(2) # restart random number generator
s1 = Session(parent_dir="/rscratch/xuanyu/data/")
n_events = 5000
count = 0
nn_list_basic = myModel.complex_cnn(9)
for hits_train, truth_train in s1.get_train_events(n=n_events, content=[s1.HITS, s1.TRUTH], randomness=True)[1]:
count += 1
print(f"{count}/{n_events}")
hits_train = join_hits_truth(hits_train, truth_train)
fy = get_target(hits_train)
loss_global = 5000
# fx = get_feature(hits, 0.0, flip=False, quadratic=True)
for i in range(100):
print("Step: " + str(i))
loss, model = train_nn(nn_list_basic, get_feature(hits_train, theta=np.random.rand() * 2 * np.pi, flip=np.random.rand() < 0.5, quadratic=True), permute_target(fy),
basic_trainable=True, epochs=5, batch_size=2048, verbose=1)
if(loss<loss_global):
print("Epoch result better than the best, saving model")
model.save("./checkpoint/aaronmao/mymodel.h5", overwrite=True)
def main():
print("start running basic neural network")
np.random.seed(1) # restart random number generator
s1 = Session(parent_dir="E:/TrackMLData/")
n_events = 50
count = 0
nn_list_basic = get_basic_nn(9)
for hits, truth in s1.get_train_events(n=n_events,
content=[s1.HITS, s1.TRUTH],
randomness=True)[1]:
count += 1
print(f"{count}/{n_events}")
hits = join_hits_truth(hits, truth)
fy = get_target(hits)
# fx = get_feature(hits, 0.0, flip=False, quadratic=True)
if count > 0:
print("validation check")
try:
train_nn(nn_list_basic,
get_feature(hits, theta=0, flip=False,
quadratic=True),
fy,
basic_trainable=False,
epochs=5,
batch_size=128,
verbose=1)
except KeyboardInterrupt:
pass
print("start actual training")
train_nn(nn_list_basic,
get_feature(hits, theta=0, flip=False, quadratic=True),
fy,
basic_trainable=True,
epochs=5,
batch_size=128,
verbose=1)
for i in range(5):
try:
train_nn(nn_list_basic,
get_feature(hits,
theta=np.random.rand() * 2 * np.pi,
flip=np.random.rand() < 0.5,
quadratic=True),
permute_target(fy),
basic_trainable=True,
epochs=5,
batch_size=128,
verbose=1)
except KeyboardInterrupt:
continue
