-
Notifications
You must be signed in to change notification settings - Fork 0
/
5_train_9_winner.py
245 lines (199 loc) · 10.4 KB
/
5_train_9_winner.py
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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
""" Libraries """
import os
import time
import json
import torch
import shutil
import random
import argparse
import numpy as np
import pandas as pd
import seaborn as sn
import matplotlib.pyplot as plt
from tqdm import tqdm
from sklearn.metrics import confusion_matrix
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from libs.dataloader import WinnerDataset
from libs.models import SingleOutputModel
from libs.misc import train_formal_list, train_informal_list
""" Classes """
class Metric():
def __init__(self, length) -> None:
self.length = length
self.values = []
def append(self, value) -> None:
self.values.append(value)
if len(self.values) > self.length: self.values.pop(0)
return
def avg(self) -> float:
return np.average(self.values)
""" Functions """
def weighted_acc(pred, truth, weight):
acc = 0.0
# assert np.max(np.max(pred), np.max(truth)) <= len(weight)
for wid, wt in enumerate(weight):
acc += (np.logical_and(pred==wid, pred==truth).sum()/np.max((1, (truth==wid).sum()))) * wt
return acc
def get_lr(optimizer) -> float:
for param_group in optimizer.param_groups:
return param_group["lr"]
def plot_confusion_matrix(confusion_matrix, filename, title):
cm_df = pd.DataFrame(confusion_matrix, index=list(range(2)), columns=list(range(2)))
plt.figure(figsize=(6, 5))
s = sn.heatmap(cm_df, annot=True)
s.set_xlabel("prediction", fontsize=10)
s.set_ylabel("truth", fontsize=10)
plt.title(title)
plt.tight_layout()
plt.savefig(filename)
plt.close()
return
def forward_step(model, batch_inputs, device) -> torch.Tensor:
batch_imgs, batch_kpts, batch_balls, batch_times, batch_bg_id, batch_hitter = batch_inputs
batch_imgs = batch_imgs.to(device)
batch_kpts = batch_kpts.to(device)
batch_balls = batch_balls.to(device)
batch_times = batch_times.to(device)
batch_bg_id = batch_bg_id.to(device)
batch_hitter = batch_hitter.to(device)
# print(batch_imgs.shape, batch_kpts.shape)
batch_pred = model(batch_imgs, batch_kpts, batch_balls, batch_times, batch_bg_id, batch_hitter)
return batch_pred
def train_epoch(model:SingleOutputModel, dataloader, criterion,
optimizer, lr_scheduler, device, weight) -> None:
loss_metric, acc_metric = Metric(50), Metric(50)
acc_weight = (np.array(weight)/sum(weight)).tolist()
confusion_matrixs = np.zeros((2, 2))
pbar = tqdm(dataloader, desc="[TRAIN]")
for batch_input, batch_truth in pbar:
model.zero_grad()
batch_pred = forward_step(model, batch_input, device)
batch_truth = batch_truth.to(device)
# print(batch_pred.shape, batch_truth.shape)
loss = criterion(batch_pred, batch_truth)
loss.backward()
optimizer.step()
lr_scheduler.step()
loss_metric.append(loss.item())
batch_truth = batch_truth.cpu().detach().numpy()
batch_pred = batch_pred.cpu().detach().numpy()
acc = weighted_acc(np.argmax(batch_pred, axis=-1), batch_truth, acc_weight)
acc_metric.append(acc)
confusion_matrixs += confusion_matrix(batch_truth, np.argmax(batch_pred, axis=-1), labels=[0,1])
pbar.set_description(f"[TRAIN] loss: {loss_metric.avg():.5f}, " + \
f"Acc: {acc_metric.avg()*100:.3f}%, " + \
f"LR: {get_lr(optimizer):.10f}")
with np.printoptions(linewidth=150, formatter={'float': '{:5.03f}'.format, 'int': '{:2} '.format}):
print("pred :", batch_pred[0], batch_pred[1], batch_pred[2], batch_pred[3], batch_pred[4])
print("pred :", np.uint8(np.argmax(batch_pred, axis=-1))[:30])
print("truth:", batch_truth[:30])
print("corct:", np.array(["X", " "])[np.uint8(np.argmax(batch_pred, axis=-1)==batch_truth)[:30]])
return loss_metric.avg(), acc_metric.avg(), confusion_matrixs
def valid_epoch(model:SingleOutputModel, dataloader,
criterion, device, weight) -> None:
loss_metric, acc_metric = Metric(1000), Metric(1000)
acc_weight = (np.array(weight)/sum(weight)).tolist()
confusion_matrixs = np.zeros((2, 2))
pbar = tqdm(dataloader, desc="[VALID]")
for batch_input, batch_truth in pbar:
model.zero_grad()
batch_pred = forward_step(model, batch_input, device)
batch_truth = batch_truth.to(device)
loss = criterion(batch_pred, batch_truth)
loss_metric.append(loss.item())
batch_truth = batch_truth.cpu().detach().numpy()
batch_pred = batch_pred.cpu().detach().numpy()
acc = weighted_acc(np.argmax(batch_pred, axis=-1), batch_truth, acc_weight)
acc_metric.append(acc)
confusion_matrixs += confusion_matrix(batch_truth, np.argmax(batch_pred, axis=-1), labels=[0,1])
pbar.set_description(f"[VALID] loss: {loss_metric.avg():.5f}, " + \
f"Acc: {acc_metric.avg()*100:.3f}%")
with np.printoptions(linewidth=150, formatter={'float': '{:5.03f}'.format, 'int': '{:2} '.format}):
print("pred :", batch_pred[0], batch_pred[1], batch_pred[2], batch_pred[3], batch_pred[4])
print("pred :", np.uint8(np.argmax(batch_pred, axis=-1))[:30])
print("truth:", batch_truth[:30])
print("corct:", np.array(["X", " "])[np.uint8(np.argmax(batch_pred, axis=-1)==batch_truth)[:30]])
return loss_metric.avg(), acc_metric.avg(), confusion_matrixs
def main(args) -> None:
os.makedirs(args.save_dir, exist_ok=True)
shutil.copy(__file__, args.save_dir)
shutil.copy("libs/models.py", args.save_dir)
shutil.copy("libs/dataloader.py", args.save_dir)
tensorboard = SummaryWriter(args.save_dir)
if args.mode == "formal": video_id_list = train_formal_list.copy()
elif args.mode == "informal": video_id_list = train_informal_list.copy()
else : video_id_list = list(range(1, 800+1))
random.shuffle(video_id_list)
train_video_id_list = video_id_list[:int(len(video_id_list)*0.8)]
valid_video_id_list = video_id_list[int(len(video_id_list)*0.8):]
print(valid_video_id_list)
with open(f"{args.save_dir}/valid_video_id_list.py", 'w') as file:
json.dump(valid_video_id_list, file, indent=4)
my_train_dataset = WinnerDataset(args.length, train_video_id_list)
my_valid_dataset = WinnerDataset(args.length, valid_video_id_list)
my_train_dataLoader = DataLoader(
my_train_dataset, args.batch_size,
shuffle=True, pin_memory=True, drop_last=True,
num_workers=args.num_workers,
)
my_valid_dataLoader = DataLoader(
my_valid_dataset, args.batch_size,
shuffle=True, pin_memory=True, drop_last=True,
num_workers=args.num_workers,
)
print("ground_truth_count:", my_train_dataset.ground_truth_count)
train_btc_avg = sum(my_train_dataset.ground_truth_count) / len(my_train_dataset.ground_truth_count)
train_weight = [ (train_btc_avg/btc) for btc in my_train_dataset.ground_truth_count ]
train_weight_torch = torch.from_numpy(np.array(train_weight)).float().to(args.device)
valid_btc_avg = sum(my_valid_dataset.ground_truth_count) / len(my_valid_dataset.ground_truth_count)
valid_weight = [ (valid_btc_avg/btc) for btc in my_valid_dataset.ground_truth_count ]
model = SingleOutputModel(length=args.length, output_dim=2, softmax=True).to(args.device)
criterion = torch.nn.CrossEntropyLoss(weight=train_weight_torch)
optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate)
lr_scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=args.lr_decay)
best_valid_loss, best_valid_acc = np.inf, 0
for epoch in range(1, args.epochs+1):
print(f"\n{epoch}/{args.epochs}")
model = model.train()
train_loss, train_acc, train_cm = \
train_epoch(model, my_train_dataLoader, criterion, optimizer, lr_scheduler, args.device, train_weight)
model = model.eval()
valid_loss, valid_acc, valid_cm = \
valid_epoch(model, my_valid_dataLoader, criterion, args.device, valid_weight)
tensorboard.add_scalar("0_Losses+LR/0_Train", train_loss, epoch)
tensorboard.add_scalar("0_Losses+LR/1_Valid", valid_loss, epoch)
tensorboard.add_scalar("0_Losses+LR/2_LR", get_lr(optimizer), epoch)
tensorboard.add_scalar("1_Accuracies/0_Train", train_acc, epoch)
tensorboard.add_scalar("1_Accuracies/1_Valid", valid_acc, epoch)
if valid_loss < best_valid_loss:
best_valid_loss = valid_loss
plot_confusion_matrix(train_cm, f"{args.save_dir}/best_valid_loss_train_cm.png", "Train Confusion Matirx at Best Valid Loss")
plot_confusion_matrix(valid_cm, f"{args.save_dir}/best_valid_loss_valid_cm.png", "Valid Confusion Matirx at Best Valid Loss")
torch.save(model, f"{args.save_dir}/best_valid_loss.pt")
if valid_acc > best_valid_acc:
best_valid_acc = valid_acc
plot_confusion_matrix(train_cm, f"{args.save_dir}/best_valid_acc_train_cm.png", "Train Confusion Matirx at Best Valid Acc")
plot_confusion_matrix(valid_cm, f"{args.save_dir}/best_valid_acc_valid_cm.png", "Valid Confusion Matirx at Best Valid Acc")
torch.save(model, f"{args.save_dir}/best_valid_acc.pt")
tensorboard.close()
return
""" Execution """
if __name__ == "__main__":
DEFAULT_MODE = "all"
DEFAULT_DEVICE = "cuda:0"
DEFAULT_SAVE_DIR = f"logs/{DEFAULT_MODE}/9_winner/{time.strftime('%Y.%m.%d-%H.%M.%S', time.localtime())}"
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--mode", type=str, default=DEFAULT_MODE)
parser.add_argument("-e", "--epochs", type=int, default=200)
parser.add_argument("-lr", "--learning-rate", type=float, default=3e-3)
parser.add_argument("-ld", "--lr-decay", type=float, default=0.998)
parser.add_argument("-bs", "--batch-size", type=int, default=64)
parser.add_argument("-nw", "--num-workers", type=int, default=4)
parser.add_argument("-l", "--length", type=int, default=151)
parser.add_argument("-d", "--device", type=str, default=DEFAULT_DEVICE)
parser.add_argument("-sd", "--save-dir", type=str, default=DEFAULT_SAVE_DIR)
args = parser.parse_args()
assert args.mode in [ "all", "formal", "informal" ]
assert (args.length-1) % 2 == 0
main(args)