def enhance_with_model(experiment_name, loadpath, cuda=False, samples=None):
checkpoint = load_to_cpu(loadpath)
p = checkpoint['p']
model = p['model_class'](**p['model_kwargs'])
model.load_state_dict(checkpoint['state_dict'])
model.transform = p['input_transform']
model.transform.target_transform = None
model.output_transform = p['output_transform']
model.inverse_transform = istft
model.experiment_name = experiment_name # this is a hack
clean_lre17_dev(model, cuda=cuda, samples=samples)
clean_lre17_eval(model, cuda=cuda, samples=samples)
clean_dataset_4_eval(model, cuda=cuda, samples=samples)
clean_lre17tel_dev(model, cuda=cuda, samples=samples)
clean_lre17tel_eval(model, cuda=cuda, samples=samples)
def clean_overfit(experiment_name, loadpath, p):
model = p['model_class'](**p['model_kwargs'])
model.load_state_dict(load_to_cpu(loadpath)['state_dict'])
model.transform = p['input_transform']
model.transform.target_transform = None
model.output_transform = p['output_transform']
model.inverse_transform = istft
model.experiment_name = experiment_name # this is a hack
samples = None # slice(-1,None,-1)
cuda = True
if cuda:
model.cuda()
clean_lre17_dev(model, cuda=cuda, samples=samples)
# clean_lre17_eval(model, cuda=cuda, samples=samples)
clean_lre17tel_dev(model, cuda=cuda, samples=samples)
def clean_overfit(experiment_name, loadpath):
from src.models.BLSTM_A5 import p
model = p['model_class'](**p['model_kwargs'])
model.load_state_dict(load_to_cpu(loadpath)['state_dict'])
model.experiment_name = experiment_name
model.transform = p['input_transform']
model.transform.mode = 'runtime'
model.apply_mask = apply_mask
model.inverse_transform = istft
samples = None # slice(-1,None,-1)
cuda = True
if cuda:
model.cuda()
clean_lre17_dev(model, cuda=cuda, samples=samples)
clean_lre17_eval(model, cuda=cuda, samples=samples)
clean_lre17tel_dev(model, cuda=cuda, samples=samples)
clean_lre17tel_eval(model, cuda=cuda, samples=samples)
clean_dataset_4_eval(model, cuda=cuda, samples=samples)
import os
from src.features.features_functions import istft
from src.evaluation.eval_BLSTM_A0 import (
clean_lre17_dev, clean_lre17_eval, clean_dataset_3_eval, clean_dataset_4_eval,
clean_lre17tel_dev, clean_lre17tel_eval, clean_lre17tel_train)
from src.models.model_functions import apply_mask, load_to_cpu
if __name__ == '__main__':
from src.models.BLSTM_A11 import p
model = p['model_class'](**p['model_kwargs'])
loadpath = os.path.join('models', 'BLSTM_A11',
'BLSTM_A11_epoch_32.state')
model.load_state_dict(load_to_cpu(loadpath)['state_dict'])
model.transform = p['input_transform']
model.transform.mode = 'runtime'
model.apply_mask = apply_mask
model.inverse_transform = istft
model.experiment_name = p['experiment_name'] # this is a hack
samples = None # slice(-1,None,-1)
cuda = True
if cuda:
model.cuda()
clean_lre17_dev(model, cuda=cuda, samples=samples)
clean_lre17_eval(model, cuda=cuda, samples=samples)
output_dir = os.path.join('data', 'interim', 'dataset_4_val',
model.experiment_name)
enhance_Datafolder(model,
input_dir,
output_dir,
batch_size=10,
cuda=cuda,
samples=samples)
if __name__ == '__main__':
# experiment_name = 'BLSTM_A5_27'
# loadpath = os.path.join('models', 'BLSTM_A5',
# 'BLSTM_A5_epoch_27.state')
# enhance_with_model(experiment_name, loadpath)
experiment_name, loadpath, cuda = False, samples = None
checkpoint = load_to_cpu(loadpath)
p = checkpoint['p']
model = p['model_class'](**p['model_kwargs'])
model.load_state_dict(checkpoint['state_dict'])
model.transform = p['input_transform']
model.transform.target_transform = None
model.output_transform = p['output_transform']
model.inverse_transform = istft
model.experiment_name = experiment_name # this is a hack
clean_dataset_4_eval(model, cuda=cuda, samples=samples)