def load_data_info(blending_data=None,
valid=None,
model_path=None,
n_hidden=256,
step=None,
beat_gt=None,
beat_subdiv=None,
model_out=".",
acoustic='kelz',
early_exit=0.001,
diagRNN=False,
ablate_list=[],
noise=None,
noise_gauss=False):
"""
Set up the global parameter dictionaries to run Bayesian Optimization with the given
settings. This is called by optimize_sk.py.
"""
global global_params
global data_dict
global model_dict
data_dict['valid'] = valid
with gzip.open(blending_data, 'rb') as file:
data_dict['blending_data'] = pickle.load(file)
if noise is None:
data_dict['noise'] = (data_dict['blending_data']['noise'] if 'noise'
in data_dict['blending_data'] else None)
data_dict['noise_gauss'] = (data_dict['blending_data']['noise_gauss']
if 'noise_gauss'
in data_dict['blending_data'] else False)
else:
data_dict['noise'] = noise
data_dict['noise_gauss'] = noise_gauss
model_param = make_model_param()
model_param['n_hidden'] = n_hidden
model_param['n_steps'] = 1 # To generate 1 step at a time
model_param['with_onsets'] = data_dict['blending_data']['with_onsets']
if diagRNN:
model_param['cell_type'] = "diagLSTM"
# Build model object
model_dict['model'] = Model(model_param)
model_dict['sess'], _ = model_dict['model'].load(model_path,
model_path=model_path)
global_params['step'] = step
global_params['beat_gt'] = beat_gt
global_params['beat_subdiv'] = beat_subdiv
global_params['model_out'] = model_out
global_params['acoustic'] = acoustic
global_params['early_exit'] = early_exit
global_params['ablate'] = ablate_list
def load_data_info(gt=None,
beam=None,
valid=None,
model_path=None,
n_hidden=256,
step=None,
model_out=".",
acoustic='kelz',
early_exit=0.001,
diagRNN=False):
global global_params
global data_dict
global model_dict
if gt is not None:
with gzip.open(gt, "rb") as file:
data_dict['gt'] = pickle.load(file)
if beam is not None:
with gzip.open(beam, "rb") as file:
data_dict['beam'] = pickle.load(file)
data_dict['valid'] = valid
model_param = make_model_param()
model_param['n_hidden'] = n_hidden
model_param['n_steps'] = 1 # To generate 1 step at a time
if diagRNN:
model_param['cell_type'] = "diagLSTM"
# Build model object
model_dict['model'] = Model(model_param)
model_dict['sess'], _ = model_dict['model'].load(model_path,
model_path=model_path)
global_params['step'] = step
global_params['model_out'] = model_out
global_params['acoustic'] = acoustic
global_params['early_exit'] = early_exit
except:
max_len = None
section = None
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
# Load model
model_param = make_model_param()
model_param['n_hidden'] = args.hidden
model_param['n_steps'] = 1 # To generate 1 step at a time
if args.diagRNN:
model_param['cell_type'] = "diagLSTM"
# Build model object
model = Model(model_param)
sess, _ = model.load(args.model, model_path=args.model)
# Load data
if args.MIDI.endswith(".mid"):
data = dataMaps.DataMaps()
data.make_from_file(args.MIDI,
args.step,
section=section,
acoustic_model=args.acoustic)
# Decode
X, Y, D = get_weight_data(data.target,
data.input,
model,
sess,