File size: 5,954 Bytes
462d118
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
import torch
from torch import nn
from torch.nn import functional as F
import torchaudio
import pytorch_lightning as pl
from torchmetrics import Accuracy, F1, Precision, Recall

import torch.nn as nn
import torch.nn.functional as F

class M11(pl.LightningModule):
    def __init__(self, hidden_units_1, hidden_units_2, dropout_1, dropout_2, n_input=1, n_output=3, stride=4, n_channel=64, lr=1e-3, l2=1e-5):
        super().__init__()
        self.save_hyperparameters()

        
        self.conv1 = nn.Conv1d(n_input, n_channel, kernel_size=80, stride=stride)
        self.bn1 = nn.BatchNorm1d(n_channel)
        self.pool1 = nn.MaxPool1d(4)

        self.conv2 = nn.Conv1d(n_channel, n_channel, kernel_size=3,padding=1)
        self.bn2 = nn.BatchNorm1d(n_channel)
        self.conv3 = nn.Conv1d(n_channel, n_channel, kernel_size=3,padding=1)
        self.bn3 = nn.BatchNorm1d(n_channel)
        self.pool2 = nn.MaxPool1d(4)

        self.conv4 = nn.Conv1d(n_channel, 2 * n_channel, kernel_size=3,padding=1)
        self.bn4 = nn.BatchNorm1d(2 * n_channel)
        self.conv5 = nn.Conv1d(2 * n_channel, 2 * n_channel, kernel_size=3,padding=1)
        self.bn5 = nn.BatchNorm1d(2 * n_channel)
        self.pool3 = nn.MaxPool1d(4)
 
        self.conv6 = nn.Conv1d(2 * n_channel, 4 * n_channel, kernel_size=3,padding=1)
        self.bn6 = nn.BatchNorm1d(4 * n_channel)
        self.conv7 = nn.Conv1d(4 * n_channel, 4 * n_channel, kernel_size=3,padding=1)
        self.bn7 = nn.BatchNorm1d(4 * n_channel)
        self.conv8 = nn.Conv1d(4 * n_channel, 4 * n_channel, kernel_size=3,padding=1)
        self.bn8 = nn.BatchNorm1d(4 * n_channel)
        self.pool4 = nn.MaxPool1d(4)

        self.conv9 = nn.Conv1d(4 * n_channel, 8 * n_channel, kernel_size=3,padding=1)
        self.bn9 = nn.BatchNorm1d(8 * n_channel)
        self.conv10 = nn.Conv1d(8 * n_channel, 8 * n_channel, kernel_size=3,padding=1)
        self.bn10 = nn.BatchNorm1d(8 * n_channel)

#         self.fc1 = nn.Linear(8 * n_channel, n_output)
        self.mlp = nn.Sequential(
            nn.Linear(8 * n_channel, hidden_units_1),
            nn.ReLU(),
            nn.Dropout(dropout_1),
            nn.Linear(hidden_units_1, hidden_units_2),
            nn.ReLU(),
            nn.Dropout(dropout_2),
            nn.Linear(hidden_units_2, n_output)
        )

    def forward(self, x):
        x = self.conv1(x)
        x = F.relu(self.bn1(x))
        x = self.pool1(x)

        x = self.conv2(x)
        x = F.relu(self.bn2(x))
        x = self.conv3(x)
        x = F.relu(self.bn3(x))
        x = self.pool2(x)

        x = self.conv4(x)
        x = F.relu(self.bn4(x))
        x = self.conv5(x)
        x = F.relu(self.bn5(x))
        x = self.pool3(x)

        x = self.conv6(x)
        x = F.relu(self.bn6(x))
        x = self.conv7(x)
        x = F.relu(self.bn7(x))
        x = self.conv8(x)
        x = F.relu(self.bn8(x))
        x = self.pool4(x)

        x = self.conv9(x)
        x = F.relu(self.bn9(x))
        x = self.conv10(x)
        x = F.relu(self.bn10(x))

        x = F.avg_pool1d(x, x.shape[-1])
        x = x.permute(0, 2, 1)
#         x = self.fc1(x)
        x = self.mlp(x)
        return F.log_softmax(x, dim=2)
    
    def training_step(self, batch, batch_idx):
        # Very simple training loop
        data, target = batch
        logits = self(data)  # this calls self.forward
        preds = torch.argmax(logits, dim=-1).squeeze()
#         loss = cost(logits.squeeze(), target)
        loss = unweighted_cost(logits.squeeze(), target)
        
        f1 = f1_metric(preds, target)
        
        self.log('train_loss', loss, on_epoch=True, prog_bar=True)
        self.log('train_f1', f1, on_epoch=True, prog_bar=True)
        return loss
    
    def validation_step(self, batch, batch_idx):
        data, target = batch
        logits = self(data)
        preds = torch.argmax(logits, dim=-1).squeeze()
#         loss = val_cost(logits.squeeze(), target)
        loss = unweighted_cost(logits.squeeze(), target)
        
        acc = accuracy(preds, target)
        f1 = f1_metric(preds, target)
        prec = precision(preds, target)
        rec = recall(preds, target)
        
        self.log('val_loss', loss, on_epoch=True, prog_bar=True)
        self.log('val_acc', acc, on_epoch=True, prog_bar=True)
        self.log('val_f1', f1, on_epoch=True, prog_bar=True)
        self.log('val_precision', prec, on_epoch=True, prog_bar=True)
        self.log('val_recall', rec, on_epoch=True, prog_bar=True)
        return loss, acc, f1, prec, rec
        
    def configure_optimizers(self):
        optimizer = torch.optim.Adam(self.parameters(), lr=self.hparams.lr, weight_decay=self.hparams.l2)
        return optimizer

# DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# model_PATH = "./model.ckpt"
# audio_PATH = "./sample_audio.wav"


# def _resample_if_necessary(signal, sr, device):
#     if sr != 8_000:
#         resampler = torchaudio.transforms.Resample(sr, 8_000).to(device)
#         signal = resampler(signal)

#     return signal

# def _mix_down_if_necessary(signal):
#     if signal.shape[0] > 1:
#         signal = torch.mean(signal, dim=0, keepdim=True)
    
#     return signal

# def get_likely_index(tensor):
#     # find most likely label index for each element in the batch
#     return tensor.argmax(dim=-1)

# model = M11.load_from_checkpoint(model_PATH).to(DEVICE)
# model.eval()

# audio, sr = torchaudio.load(audio_PATH)
# # resampler = torchaudio.transforms.Resample(sr, 8_000).to(DEVICE)
# processed_audio = _mix_down_if_necessary(_resample_if_necessary(audio, sr, DEVICE))

# print(processed_audio.shape)
# with torch.no_grad():
#     pred = get_likely_index(model(processed_audio.unsqueeze(0).to(DEVICE))).view(-1)
    
# # y_true = target.tolist()
# # y_pred = pred.tolist()
# # target_names = eval_dataset.label_list
# # print(classification_report(y_true, y_pred, target_names=target_names))
# print(pred)