File size: 9,681 Bytes
29d330e | 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 | from easyfsl.methods import PrototypicalNetworks, FewShotClassifier
from typing import Any
from torch.optim import SGD, Optimizer
from tqdm import tqdm
import numpy as np
import torch
from statistics import mean
from torch.optim.lr_scheduler import MultiStepLR
from easyfsl.utils import evaluate
import os
def mmd(x, y, mode=0):
if mode == 0:
diff = torch.mean(x, dim=0) - torch.mean(y, dim=0)
else:
xf = x.flatten()
yf = y.flatten()
diff = xf - yf
return diff #torch.sum(diff**2)
def training_epoch(
model: FewShotClassifier,
data_loader: Any,
optimizer: Optimizer,
epoch_: int,
round_name: int,
station_no: int,
weighted_loss: bool = False,
embedding_diff: bool = False,
track_emb_loss_epoch: bool = False,
globel_model_1_2: Any = None,
embedding_diff_plus_loss: bool = False,
emb_diff_to_server: bool = False,
loss_funct: str = None,
LOG_FILE_emb = Any
):
emb_df = 0
emb_list = []
all_loss = []
model.train()
LOSS_FUNCTION = torch.nn.CrossEntropyLoss()
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
similar_classes = {
0: [0,1, 3, 5, 7],
1: [0,1, 2, 4, 5],
2: [1,2],
3: [0, 3,4, 6],
4: [1, 3,4],
5: [0, 1,5],
6: [3,6],
7: [0,7, 8],
8: [7,8, 10],
9: [9,10],
10: [8, 9,10],
}
similarity_matrix = 0.75 * torch.ones((11, 11)).to(DEVICE)
# Update the matrix with 1 for similar classes
for class_idx, similar_list in similar_classes.items():
similarity_matrix[class_idx, similar_list] = 1.0
with tqdm(
enumerate(data_loader), total=len(data_loader), desc="Training"
) as tqdm_train:
for episode_index, (support_images, support_labels, query_images, query_labels, _) in tqdm_train:
optimizer.zero_grad()
print("im under training epoch-----------------------------")
model.process_support_set(support_images.to(DEVICE), support_labels.to(DEVICE))
classification_scores = model(query_images.to(DEVICE))
weight_matrix = similarity_matrix[np.ix_([0,1,2,3,4,5,7,10],[0,1,2,3,4,5,7,10])]
weight_matrix = weight_matrix.T
weight_matrix = weight_matrix[query_labels]
print(weight_matrix.shape)
if weighted_loss:
print("Im Under weighted_loss")
classification_scores = classification_scores * weight_matrix
if embedding_diff:
print(f"Loss function is based on embedding difference on seen classes")
globel_model_1_2.process_support_set(support_images.to(DEVICE), support_labels.to(DEVICE))
classification_scores_ref = globel_model_1_2(query_images.to(DEVICE))
# print("im calculating embb distance")
# print(classification_scores_ref.shape)
# print(classification_scores.shape)
if loss_funct == "MSE":
print("loss function is MSE")
# Assuming classification_scores_ref and classification_scores are tensors
emb_diff = torch.mean((classification_scores_ref - classification_scores)**2, dim=0)
elif loss_funct == "MMD":
print("loss function is MMD")
emb_diff = mmd(classification_scores_ref,classification_scores)
else:
print("loss function is Absolute difference")
emb_diff = torch.mean(abs(classification_scores_ref - classification_scores))
emb_diff = emb_diff
"""""you can save the embedding difference in a file"""
# if epoch_ == 4 or epoch_ ==1 or epoch_ == 0:
# # combined_tensor = torch.stack([classification_scores, classification_scores_ref])
# array_emb = emb_diff.clone().detach().cpu().numpy()
# # array_emb = combined_tensor.detach().cpu().numpy()
# print("Im appending the matrix--------------again and again")
# log_npy_path = LOG_FILE_emb + f"_{round_name}.npy"
# with open(log_npy_path, "ab") as file:
# np.save(file,array_emb)
""""choose the station number to set the embedding difference to zero"""
# # if epoch_ == 4:
# # with open(LOG_FILE_emb, "a") as file:
# # file.write(f"emb_diff: {emb_diff}\n")
# if station_no == 0:
# emb_diff[:3] = 0
# elif station_no == 1:
# emb_diff[:3] = 0
# else:
# emb_diff[:3] = 0
# # classification_scores += emb_diff
emb_diff[:3] = 0
if embedding_diff_plus_loss:
print(f"Loss function is based on embedding difference on seen classes and loss value")
globel_model_1_2.process_support_set(support_images.to(DEVICE), support_labels.to(DEVICE))
classification_scores_ref = globel_model_1_2(query_images.to(DEVICE))
emb_diff = classification_scores_ref - classification_scores
emb_diff = abs(emb_diff)
emb_diff = torch.sum(emb_diff)
emb_diff = emb_diff/100
if emb_diff_to_server:
globel_model_1_2.process_support_set(support_images.to(DEVICE), support_labels.to(DEVICE))
classification_scores_ref = globel_model_1_2(query_images.to(DEVICE))
emb_diff = classification_scores_ref - classification_scores
sum_emb = torch.sum(emb_diff).item()
emb_df += sum_emb
loss = LOSS_FUNCTION(classification_scores, query_labels.to(DEVICE))
if embedding_diff:
if loss_funct == "MMD":
emb_loss_epoch = torch.sum(emb_diff**2)*0.1
# loss += emb_loss_epoch
else:
emb_loss_epoch = torch.sum(emb_diff)*0.1
# loss += emb_loss_epoch
if embedding_diff_plus_loss:
print(loss)
loss+= emb_diff
print(loss)
print(loss)
loss.backward()
optimizer.step()
all_loss.append(loss.item())
tqdm_train.set_postfix(loss=mean(all_loss))
if emb_diff_to_server:
print("Im using emb_diff_to_server to return loss" )
return mean(all_loss), emb_df
else:
return mean(all_loss), emb_loss_epoch
def train(remote_train_dataset,
station_model,
n_epochs,
learning_rate,
model,
scheduler_milestones,
scheduler_gamma,
LOG_FILE,
local_model,
round,
LOG_FILE_emb,
global_model,
station_n,
ls_function,
embedding_diff = False,
emb_diff_to_server =False,
embedding_diff_plus_loss=False,
):
emb_df_accumulated =0
ret_val = None
for epoch in range(n_epochs):
print("Im Under train")
station_model = station_model
station_optimizer = SGD(model.parameters(), lr=learning_rate, momentum=0.9, weight_decay=5e-4)
scheduler = MultiStepLR(station_optimizer, milestones=scheduler_milestones, gamma=scheduler_gamma)
print(f"Epoch for station 1: {epoch}")
average_loss = training_epoch(station_model, remote_train_dataset, station_optimizer,embedding_diff=embedding_diff,embedding_diff_plus_loss=embedding_diff_plus_loss,emb_diff_to_server=emb_diff_to_server, globel_model_1_2=global_model,LOG_FILE_emb=LOG_FILE_emb,epoch_=epoch,round_name=round,station_no=station_n,loss_funct=ls_function)
if emb_diff_to_server:
print(f"average_loss is {average_loss}")
embb_difference = average_loss[1]
emb_df_accumulated +=embb_difference
ret_val = average_loss[1]
# Log loss to file
with open(LOG_FILE, "a") as file:
file.write(f"Epoch: {epoch}, Loss: {average_loss[0]},Embedding loss: {average_loss[1]}\n")
scheduler.step()
print(f"local model path is ================{local_model}=============")
model_file = os.path.join(local_model, f"round--{round}_weights.h5")
torch.save(station_model.state_dict(), model_file)
if emb_diff_to_server:
return emb_df_accumulated
else:
return ret_val.clone().detach().cpu().numpy() ##added
def test(net, remote_valid_dataset,LOG_FILE,DEVICE):
"""Evaluate the network on the entire test set."""
print("Im Under test")
net.eval()
validation_accuracy = evaluate(net, remote_valid_dataset, device=DEVICE, tqdm_prefix="Validation")
with open(LOG_FILE, "a") as file:
file.write(f"accuracy: {validation_accuracy}\n")
return validation_accuracy |