tiny-FWFM/pruned_federated_eval_radio_sig_identification.py at main · Mohammad-Hallaq/tiny-FWFM · 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
from tqdm import tqdm

from torch.utils.data import DataLoader

from dataset_classes.radio_sig import RadioSignal
import torch
import models_vit
import numpy as np
from sklearn.model_selection import StratifiedShuffleSplit
from sklearn.metrics import confusion_matrix
import random
from pathlib import Path
import matplotlib.pyplot as plt
import seaborn as sns
plt.rcParams['font.family'] = 'serif'


import torch.nn.functional as F
import timm

def forward(self, x):
    """https://github.com/huggingface/pytorch-image-models/blob/054c763fcaa7d241564439ae05fbe919ed85e614/timm/models/vision_transformer.py#L79"""
    B, N, C = x.shape
    qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
    q, k, v = qkv.unbind(0)
    q, k = self.q_norm(q), self.k_norm(k)

    if self.fused_attn:
        x = F.scaled_dot_product_attention(
            q, k, v,
            dropout_p=self.attn_drop.p,
        )
    else:
        q = q * self.scale
        attn = q @ k.transpose(-2, -1)
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        x = attn @ v

    x = x.transpose(1, 2).reshape(B, N, -1) # original implementation: x = x.transpose(1, 2).reshape(B, N, C)
    x = self.proj(x)
    x = self.proj_drop(x)
    return x

seed = 42
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
# dataset = RadioSignal(Path('../datasets/radio_sig_identification'))
dataset_train = RadioSignal(Path('fine-tuning_datasets/radio_sig_identification/train'))
dataset_test = RadioSignal(Path('fine-tuning_datasets/radio_sig_identification/test'))

# splitter = StratifiedShuffleSplit(n_splits=1, train_size=0.8, test_size=0.2, random_state=seed)
# all_labels = [dataset[i][1] for i in range(len(dataset))]
#
# for train_idx, test_idx in splitter.split(range(len(dataset)), all_labels):
#     dataset_train = torch.utils.data.Subset(dataset, train_idx)
#     dataset_test = torch.utils.data.Subset(dataset, test_idx)

dataloader_train = DataLoader(dataset_train, batch_size=256, shuffle=True, num_workers=0)
dataloader_test = DataLoader(dataset_test, batch_size=256, shuffle=False, num_workers=0)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = models_vit.__dict__['vit_small_patch16'](global_pool='token', num_classes=10, drop_path_rate=0.1, in_chans=1, head_layers=1)
# print(model)
# checkpoint = torch.load(Path('/home/ict317-3/Mohammad/mae/output_dir/checkpoint-190.pth'), map_location='cpu', weights_only=False)
# msg = model.load_state_dict(checkpoint['model'], strict=True)
# print(msg)

model = torch.load('/home/ict317-3/Mohammad/mae/sig_id_output_dir/fed_avg_model.pth', weights_only=False)

print(model)

for m in model.modules():
            if isinstance(m, timm.models.vision_transformer.Attention):
                m.forward = forward.__get__(m, timm.models.vision_transformer.Attention)


model = model.to(device)
model.eval()

class_names = ['ais', 'bluetooth', 'cellular', 'fm', 'lora','packet', 'rke', 'RS41-Radiosonde', 'sstv', 'wifi']
all_labels_train = []
all_preds_train = []

with torch.no_grad():
    for samples, targets in tqdm(dataloader_train, desc='Train batch'):
        samples, targets = samples.to(device), targets.to(device)
        output = model(samples)
        all_preds_train.extend(output.argmax(dim=-1).cpu().numpy())
        all_labels_train.extend(targets.cpu().numpy())

all_labels_test = []
all_preds_test = []

with torch.no_grad():
    for samples, targets in tqdm(dataloader_test, desc='Test batch'):
        samples, targets = samples.to(device), targets.to(device)
        output = model(samples)
        all_preds_test.extend(output.argmax(dim=-1).cpu().numpy())
        all_labels_test.extend(targets.cpu().numpy())

conf_mat_train = confusion_matrix(all_labels_train, all_preds_train)
conf_mat_test = confusion_matrix(all_labels_test, all_preds_test)


conf_mat_train = conf_mat_train.astype(np.float32)
conf_mat_test = conf_mat_test.astype(np.float32)
for i in range(len(class_names)):
    conf_mat_train[i] /= np.sum(conf_mat_train[i])
    conf_mat_test[i] /= np.sum(conf_mat_test[i])

accuracy_train = np.mean(np.diagonal(conf_mat_train))
accuracy_test = np.mean(np.diagonal(conf_mat_test))

fig, axs = plt.subplots(1, 2, figsize=(10, 5))
fig.suptitle('Finetuning frozen ViT-S + 2 Layer Head')
sns.heatmap(conf_mat_train, cmap='Reds', yticklabels=class_names, ax=axs[0])
sns.heatmap(conf_mat_test, cmap='Reds',  yticklabels=class_names, ax=axs[1])
axs[0].tick_params(axis='y', labelsize=10)
axs[1].tick_params(axis='y', labelsize=10)
axs[0].tick_params(axis='x', which='both', bottom=False, top=False, labelbottom=False)
axs[1].tick_params(axis='x', which='both', bottom=False, top=False, labelbottom=False)
axs[0].set_title(f'Train - Accuracy: {accuracy_train:.2f}')
axs[1].set_title(f'Test - Accuracy: {accuracy_test:.2f}')
plt.tight_layout()
plt.savefig('/home/ict317-3/Mohammad/mae/sig_id_output_dir/conf_mat_radio_identification.png', dpi=400)
plt.show()