def __init__(self,
model,
optimizer,
loss_func,
train_loader,
val_loader=None,
scheduler=None,
config=None):
super().__init__(model,
optimizer,
loss_func,
train_loader,
val_loader=val_loader,
scheduler=scheduler,
config=config)
self.enc = fb.Encoder(fb.STFTFB(**config['filterbank']))
def make_model_and_optimizer(conf):
""" Function to define the model and optimizer for a config dictionary.
Args:
conf: Dictionary containing the output of hierachical argparse.
Returns:
model, optimizer.
The main goal of this function is to make reloading for resuming
and evaluation very simple.
"""
enc = fb.Encoder(fb.STFTFB(**conf['filterbank']))
masker = ChimeraPP(int(enc.filterbank.n_feats_out/2), 2,
embedding_dim=20, n_layers=2, hidden_size=600, \
dropout=0, bidirectional=True)
model = Model(enc, masker)
optimizer = make_optimizer(model.parameters(), **conf['optim'])
return model, optimizer
def make_encoder_from(fb_class, config):
enc = fb.Encoder(fb_class(**config))
fb_dim = enc.filterbank.n_feats_out
return enc, fb_dim
import numpy as np
import torch
from torch.utils.data import DataLoader
import torch.nn as nn
from torch import optim
from asteroid.filterbanks.transforms import take_mag
import asteroid.filterbanks as fb
from asteroid.data.wsj0_mix import WSJ2mixDataset, BucketingSampler, \
collate_fn
from asteroid.masknn.blocks import SingleRNN
from asteroid.losses import PITLossWrapper, pairwise_mse
from asteroid.losses import deep_clustering_loss
EPS = torch.finfo(torch.float32).eps
enc = fb.Encoder(fb.STFTFB(256, 256, stride=64))
enc = enc.cuda()
parser = argparse.ArgumentParser()
parser.add_argument('--gpus', type=str, help='list of GPUs', default='-1')
parser.add_argument('--exp_dir',
default='exp/tmp',
help='Full path to save best validation model')
pit_loss = PITLossWrapper(pairwise_mse, mode='pairwise')
class Model(nn.Module):
def __init__(self):
#def __init__(self, in_chan, n_src, rnn_type = 'lstm',
# embedding_dim=20, n_layers=2, hidden_size=600,