示例#1
0

#--------------------------------------------
# Dataset-specific constants and functions + loading
#--------------------------------------------

# Useful Constants

# Those are separate normalised input features for the neural network

# Load "X" (the neural network's training and testing inputs)
X_train_signals_paths = "/localSSD/xjc/codalab_train/train/train_train_fc7_feature_new.fea"
X_test_signals_paths = "/localSSD/xjc/codalab_train/train/train_valid_fc7_feature_new.fea"
y_train_path = "/localSSD/xjc/codalab_train/train/train_y_label_2.fea"
y_test_path = "/localSSD/xjc/codalab_train/train/valid_y_label_2.fea"
X_train = load_X_pca(X_train_signals_paths)
X_test = load_X_pca(X_test_signals_paths)
X_train = do_pca(X_train)
X_test = do_pca(X_test)
y_train = one_hot(load_Y_my(y_train_path))
y_test = one_hot(load_Y_my(y_test_path))
#--------------------------------------------
# Training (maybe multiple) experiment(s)
#--------------------------------------------

n_layers_in_highway = 0
n_stacked_layers = 3
trial_name = "{}x{}".format(n_layers_in_highway, n_stacked_layers)

for learning_rate in [0.0001]:  # [0.01, 0.007, 0.001, 0.0007, 0.0001]:
    for lambda_loss_amount in [0.005]:
示例#2
0
        # NOTE: values of exactly 1 (int) for those 2 high-level parameters below totally disables them and result in only 1 starting LSTM.
        self.n_layers_in_highway = 3  # Number of residual connections to the LSTMs (highway-style), this is did for each stacked block (inside them).
        self.n_stacked_layers = 3  # Stack multiple blocks of residual
        # layers.


#--------------------------------------------
# Dataset-specific constants and functions + loading
#--------------------------------------------

# Useful Constants

# Those are separate normalised input features for the neural network

X_test_path = "path/to/feature"
X_test_result = load_X_pca(X_test_path)
X_test_result = do_pca(X_test_result)

n_layers_in_highway = 0
n_stacked_layers = 2
trial_name = "{}x{}".format(n_layers_in_highway, n_stacked_layers)


class EditedConfig(Config):
    def __init__(self, X, Y):
        super(EditedConfig, self).__init__(X, Y)
        self.n_layers_in_highway = n_layers_in_highway
        self.n_stacked_layers = n_stacked_layers


pred_out = test_with_config(EditedConfig, X_test_result)
示例#3
0
#X_test = load_X_my(X_test_signals_paths)

X_train_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_our_face_correct/vgg16_afewfacetrain_fc6.fea"
#X_test_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_our_face_correct/vgg16_afewfaceval_fc6.fea"
#y_train_path = "/home/xujinchang/share/caffe-center-loss/correct_lstm/lstm_train_label_correct.txt"
#y_test_path = "/home/xujinchang/share/caffe-center-loss/correct_lstm/lstm_val_label_correct.txt"
#X_train_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_afew_face/lbp_vgg16_face_train_fc6.fea"
#X_test_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_afew_face/lbp_vgg16_face_val_fc6.fea"
# X_train_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_our_resnet/res34_afewfacetrain.fea"
# X_test_signals_paths = "/home/xujinchang/share/caffe-center-loss/lstm_our_resnet/res34_afewfaceval.fea"
y_train_path = "/home/xujinchang/share/caffe-center-loss/lstm_afew_face/lbp_afewface_train_label.fea"
y_test_path = "/home/xujinchang/share/caffe-center-loss/lstm_afew_face/lbp_afewface_val_label.fea"
#y_test_path ="/home/xujinchang/share/caffe-center-loss/lstm_afew_face/our_test_fake_label"
X_test_signals_paths = '/home/xujinchang/share/caffe-center-loss/lstm_afew_face/our_vgg16_face_test_fc6.fea'

X_train_ori = load_X_pca(X_train_signals_paths)
X_test_ori = load_X_pca(X_test_signals_paths)
standard = StandardScaler()
X_train_norm = standard.fit_transform(X_train_ori)
X_test_norm = standard.transform(X_test_ori)
pca = PCA(n_components=1024)
X_train = pca.fit_transform(X_train_norm)
X_test = pca.transform(X_test_norm)

#X_train = reshape_pca(X_train_norm)
X_valid_result = reshape_pca(X_test)
#y_train = one_hot(load_Y_my(y_train_path))
y_test = one_hot(load_Y_my(y_test_path))

#X_valid_result = load_X_pca(X_valid_path)
#X_valid_result = do_pca(X_valid_result)
示例#4
0
# Dataset-specific constants and functions + loading
#--------------------------------------------

# Useful Constants

# Those are separate normalised input features for the neural network

X_train_signals_paths = "/home/xujinchang/caffe-blur-pose/valid_fc7_feature_new.fea"
X_test_signals_paths = "/home/xujinchang/caffe-blur-pose/test_fc7_feature_new.fea"
y_train_path = "/home/xujinchang/caffe-blur-pose/valid_y_label_2.fea"
y_test_path = "/home/xujinchang/caffe-blur-pose/valid_y_label_2.fea"
#X_train = load_X_my(X_train_signals_paths)
X_valid_path = "/localSSD/xjc/codalab_train/valid/valid_fc7_feature_new.fea"
#X_valid_path = "/localSSD/xjc/codalab_train/test/final_fc7_feature_new.fea"
#X_test = load_X_my(X_test_signals_paths)
X_valid_result = load_X_pca(X_valid_path)
X_valid_result = do_pca(X_valid_result)

n_layers_in_highway = 0
n_stacked_layers = 2
trial_name = "{}x{}".format(n_layers_in_highway, n_stacked_layers)


class EditedConfig(Config):
    def __init__(self, X, Y):
        super(EditedConfig, self).__init__(X, Y)
        self.n_layers_in_highway = n_layers_in_highway
        self.n_stacked_layers = n_stacked_layers


pred_out = test_with_config(EditedConfig, X_valid_result)