import hat.backend as K
import config as cfg
import prepare_data as pp_data
# reshape image from N*1024 to N*3*32*32
def reshape_img_for_cnn(x):
N = x.shape[0]
return np.reshape(x, (N, 3, 32, 32))
# load data
tr_X, tr_y, te_X, te_y = pp_data.load_data()
# normalize data
scaler = pp_data.get_scaler(tr_X)
tr_X = pp_data.transform(tr_X, scaler)
te_X = pp_data.transform(te_X, scaler)
# reshape X to shape: (n_pictures, n_fmaps=3, n_row=32, n_col=32)
tr_X = reshape_img_for_cnn(tr_X)
te_X = reshape_img_for_cnn(te_X)
# init params
n_out = 10
# sparse label to 1-of-K categorical label
tr_y = sparse_to_categorical(tr_y, n_out)
te_y = sparse_to_categorical(te_y, n_out)
print tr_X.shape
print tr_y.shape
optimizer=optimizer,
callbacks=callbacks)
if __name__ == '__main__':
# hyper-params
n_concat = 11 # concatenate frames
hop = 5 # step_len
fold = 0 # can be 0, 1, 2, 3
dev_fe_fd = cfg.dev_fe_logmel_fd
eva_fd_fd = cfg.eva_fe_logmel_fd
if sys.argv[1] == "--dev_train":
scaler = pp_data.get_scaler(fe_fd=dev_fe_fd,
csv_file=cfg.dev_tr_csv[fold],
with_mean=True,
with_std=True)
train(tr_fe_fd=dev_fe_fd,
tr_csv_file=cfg.dev_tr_csv[fold],
te_fe_fd=dev_fe_fd,
te_csv_file=cfg.dev_te_csv[fold],
n_concat=n_concat,
hop=hop,
scaler=scaler,
out_md_fd=cfg.dev_md_fd)
elif sys.argv[1] == "--dev_recognize":
scaler = pp_data.get_scaler(fe_fd=dev_fe_fd,
csv_file=cfg.dev_tr_csv[fold],
with_mean=True,
with_std=True)
hop=hop)
if __name__ == '__main__':
# hyper-params
n_concat = cfg.n_concat # concatenate frames
hop = cfg.hop # step_len
fold = 0 # can be 0, 1, 2, 3
# your workspace
dev_feature = cfg.dev_mel
eva_feature = cfg.eva_mel
if sys.argv[1] == "--all":
scaler = pp_data.get_scaler(fe_fd=dev_feature,
csv_file=cfg.dev_tr[fold],
with_mean=True,
with_std=True)
dev_train()
scaler = pp_data.get_scaler(fe_fd=dev_feature,
csv_file=cfg.dev_meta,
with_mean=True,
with_std=True)
eva_train()
elif sys.argv[1] == "--dev_train":
scaler = pp_data.get_scaler(fe_fd=dev_feature,
csv_file=cfg.dev_tr[fold],
with_mean=True,
with_std=True)
#scaler = joblib.load(cfg.ld_sc)
callbacks=callbacks)
if __name__ == '__main__':
# hyper-params
n_concat = 11 # concatenate frames
hop = 5 # step_len
fold = 0 # can be 0, 1, 2, 3
dev_fe_fd = cfg.dev_fe_logmel_fd
eva_fd_fd = cfg.eva_fe_logmel_fd
if sys.argv[1] == "--dev_train":
scaler = pp_data.get_scaler(fe_fd=dev_fe_fd,
csv_file=cfg.dev_tr_csv[fold],
with_mean=True,
with_std=True)
train(tr_fe_fd=dev_fe_fd,
tr_csv_file=cfg.dev_tr_csv[fold],
te_fe_fd=dev_fe_fd,
te_csv_file=cfg.dev_te_csv[fold],
n_concat=n_concat,
hop=hop,
scaler=scaler,
out_md_fd=cfg.dev_md_fd)
elif sys.argv[1] == "--dev_recognize":
scaler = pp_data.get_scaler(fe_fd=dev_fe_fd,
csv_file=cfg.dev_tr_csv[fold],
with_mean=True,
with_std=True)