paperSoundscapeSourceClassification/trainers.py at main · modantailleur/paperSoundscapeSourceClassification · GitHub

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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Sep 27 17:45:28 2022

@author: user
"""

import torch
import torch.optim as optim
import torch.utils.data
from tqdm import tqdm
import numpy as np
import torch.nn as nn
import copy
from sklearn.model_selection import KFold
from dataloader import Dataset
import models as md

#for CNN + PINV
class DLModelsTrainer:
    """
    Trainer to train the coarser (527 --> 3)
    """
    def __init__(self, models_path, scores, groundtruth, fname, scores_len, learning_rate=1e-3, dtype=torch.FloatTensor,
                 ltype=torch.LongTensor):

        self.dtype = dtype
        self.ltype = ltype

        self.models_path = models_path

        k_folds = 10
        self.kfold = KFold(n_splits=k_folds, shuffle=True, random_state=0)
        self.dataset_fold = self.kfold.split(fname)

        self.scores_len = scores_len
        self.model = md.FC(scores_len)
        self.learning_rate = learning_rate
        self.optimizer = optim.Adam(params=self.model.parameters(), lr=learning_rate)
        self.model_fold = []

        self.scores = scores
        self.groundtruth = groundtruth
        self.fname = fname
        self.train_fold = []
        self.eval_fold = []

    def train(self, batch_size=128, epochs=100, device=torch.device("cpu")):
        losses_train_fold = []
        losses_eval_fold = []

        for fold, (train_id, eval_id) in enumerate(self.dataset_fold):
            self.train_fold.append(train_id)
            self.eval_fold.append(eval_id)

            train_scores = self.scores[train_id]
            train_groundtruth = self.groundtruth[train_id]
            eval_scores = self.scores[eval_id]
            eval_groundtruth = self.groundtruth[eval_id]

            model = copy.deepcopy(self.model)
            optimizer = optim.Adam(params=model.parameters(), lr=self.learning_rate)
            loss_function = nn.BCELoss()
            model = model.to(device)

            cur_loss = 0

            train_dataset = Dataset(train_scores, train_groundtruth)
            train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=8, pin_memory=False, drop_last=False)

            losses_train = []
            losses_eval = []
            #train
            for epoch in range(epochs):
                tqdm_it=tqdm(train_dataloader,
                            desc='TRAINING: Epoch {}, loss: {:.4f}'
                            .format(epoch+1, cur_loss))
                for (x,y) in tqdm_it:
                    x = x.type(self.dtype)
                    y = y.type(self.dtype)

                    x = x.to(device)
                    y = y.to(device)

                    optimizer.zero_grad()

                    y_pred = model(x)

                    loss = loss_function(y_pred,y)

                    loss.backward()

                    optimizer.step()

                    cur_loss = float(loss.data)

                    losses_train.append(cur_loss)

                    tqdm_it.set_description('TRAINING: Epoch {}, loss: {:.4f}'
                                            .format(epoch+1,cur_loss))

                loss_eval = self.validate(eval_scores, eval_groundtruth, model, epoch+1, batch_size=batch_size, device=device, label='EVALUATION')
                losses_eval.append(loss_eval)

            losses_train_fold.append(losses_train)
            losses_eval_fold.append(losses_eval)
            self.model_fold.append(model.state_dict())

        return(losses_train_fold, losses_eval_fold)

    def validate(self, eval_scores, eval_groundtruth, model, currentEpoch, batch_size=64, device=torch.device("cpu"), label='VALIDATION', forced=False,
                    n_chunk=100):

        loss_valid = 0
        len_valid_dataset = 0
        loss_function =  nn.BCELoss(reduction='none')
        all_losses = torch.Tensor().to(device)

        valid_dataset = Dataset(eval_scores, eval_groundtruth)
        valid_dataloader = torch.utils.data.DataLoader(valid_dataset, batch_size=batch_size, shuffle=True, num_workers=8, pin_memory=False, drop_last=False)
        tqdm_it=tqdm(valid_dataloader, desc=label+':loss {:.4f}'.format(0))
        for (x,y) in tqdm_it:
            x = x.type(self.dtype)
            y = y.type(self.dtype)

            x = x.to(device)
            y = y.to(device)

            y_pred = model(x)

            loss = loss_function(y_pred,y).mean(dim=[1]).detach()
            all_losses = torch.cat((all_losses,loss))

        loss_valid = torch.mean(all_losses)
        tqdm_it.set_description(label+':loss {:.4f}'.format(loss_valid))
        print(f'loss_valid:{loss_valid}')
        return loss_valid.detach().cpu().numpy()