-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathtrain.py
159 lines (140 loc) · 6.2 KB
/
train.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
import torch
import torch.nn.functional as F
from torch.nn import Linear
import time
from torch import tensor
import torch.nn
from utils import TSPLoss
import pickle
from torch.utils.data import Dataset,DataLoader# use pytorch dataloader
from random import shuffle
import numpy as np
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--seed', type=int, default=42, help='Random seed.')
parser.add_argument('--num_of_nodes', type=int, default=200, help='Graph Size')
parser.add_argument('--lr', type=float, default=3e-3,
help='Learning Rate')
parser.add_argument('--moment', type=int, default=1,
help='scattering moment')
parser.add_argument('--hidden', type=int, default=64,
help='Number of hidden units.')
parser.add_argument('--batch_size', type=int, default=32,
help='batch_size')
parser.add_argument('--nlayers', type=int, default=3,
help='num of layers')
parser.add_argument('--EPOCHS', type=int, default=20,
help='epochs to train')
parser.add_argument('--penalty_coefficient', type=float, default=2.,
help='penalty_coefficient')
parser.add_argument('--wdecay', type=float, default=0.0,
help='weight decay')
parser.add_argument('--temperature', type=float, default=3.5,
help='temperature for adj matrix')
parser.add_argument('--rescale', type=float, default=2.,
help='rescale for xy plane')
parser.add_argument('--C1_penalty', type=float, default=10.,
help='penalty row/column')
parser.add_argument('--topk', type=int, default=10,
help='topk')
parser.add_argument('--stepsize', type=int, default=20,
help='step size')
parser.add_argument('--diag_loss', type=float, default=0.1,
help='penalty on the diag')
args = parser.parse_args()
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.enabled = True
torch.cuda.manual_seed(args.seed)
np.random.seed(args.seed)
### load train instance
tsp_instances = np.load('./data/train_tsp_instance_%d.npy'%args.num_of_nodes)
NumofTestSample = tsp_instances.shape[0]
Std = np.std(tsp_instances, axis=1)
Mean = np.mean(tsp_instances, axis=1)
tsp_instances = tsp_instances - Mean.reshape((NumofTestSample,1,2))
#tsp_instances = np.divide(tsp_instances,Std.reshape((NumofTestSample,1,2)))
tsp_instances = args.rescale * tsp_instances # 2.0 is the rescale
tsp_sols = np.load('./data/train_tsp_sol_%d.npy'%args.num_of_nodes)
dataset_scale = 1
LENGDATA = tsp_instances.shape[0]
total_samples = int(np.floor(LENGDATA*dataset_scale))
import json
preposs_time = time.time()
from models import GNN
#scattering model
model = GNN(input_dim=2, hidden_dim=args.hidden, output_dim=args.num_of_nodes, n_layers=args.nlayers)
from scipy.spatial import distance_matrix
### count model parameters
def count_parameters(model):
return sum(p.numel() for p in model.parameters() if p.requires_grad)
print('Total number of parameters:')
print(count_parameters(model))
#dis_mat = distance_matrix(tsp_instances[0],tsp_instances[0])
# https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance_matrix.html
def coord_to_adj(coord_arr):
dis_mat = distance_matrix(coord_arr,coord_arr)
return dis_mat
tsp_instances_adj = np.zeros((LENGDATA,args.num_of_nodes,args.num_of_nodes))
for i in range(LENGDATA):
tsp_instances_adj[i] = coord_to_adj(tsp_instances[i])
#print(coord_to_adj(tsp_instances[0]))
class TSP_Dataset(Dataset):
def __init__(self, coord,data, targets):
self.coord = torch.FloatTensor(coord)
self.data = torch.FloatTensor(data)
self.targets = torch.LongTensor(targets)
def __getitem__(self, index):
xy_pos = self.coord[index]
x = self.data[index]
y = self.targets[index]
# tsp_instance = Data(coord=x,sol=y)
return tuple(zip(xy_pos,x,y))
def __len__(self):
return len(self.data)
dataset = TSP_Dataset(tsp_instances,tsp_instances_adj,tsp_sols)
#num_trainpoints = int(np.floor(0.6*total_samples))
num_trainpoints = 2000
num_valpoints = total_samples - num_trainpoints
#num_testpoints = total_samples - (num_trainpoints + num_valpoints)
sctdataset = dataset
traindata= sctdataset[0:num_trainpoints]
#traindata = sctdataset[0:1000]
#valdata = sctdataset[num_trainpoints:]
valdata = sctdataset[num_trainpoints:]
batch_size = args.batch_size
train_loader = DataLoader(traindata, batch_size, shuffle=True)
val_loader = DataLoader(valdata, batch_size, shuffle=False)
from torch.optim.lr_scheduler import StepLR
#optimizer = torch.optim.RMSprop(model.parameters(), lr=args.lr,weight_decay=args.wdecay)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr,weight_decay=args.wdecay)
scheduler = StepLR(optimizer, step_size=args.stepsize, gamma=0.8)
model.cuda()
mask = torch.ones(args.num_of_nodes, args.num_of_nodes).cuda()
mask.fill_diagonal_(0)
def train(epoch):
scheduler.step()
model.train()
print('Epoch: %d'%epoch)
for batch in train_loader:
f0 = batch[0].cuda()
distance_m = batch[1].cuda()
adj = torch.exp(-1.*distance_m/args.temperature)
adj *= mask
output = model(f0,adj)
TSPLoss_constaint,Heat_mat = TSPLoss(SctOutput=output,distance_matrix=distance_m,num_of_nodes=args.num_of_nodes)
Heat_mat_diagonals = [torch.diagonal(mat) for mat in Heat_mat]
Heat_mat_diagonals = torch.stack(Heat_mat_diagonals, dim=0)
Nrmlzd_constraint = (1. - torch.sum(output,2))**2
Nrmlzd_constraint = torch.sum(Nrmlzd_constraint)
loss = args.C1_penalty * Nrmlzd_constraint + 1.*torch.sum(TSPLoss_constaint) + args.diag_loss*torch.sum(Heat_mat_diagonals)
batchloss = torch.sum(loss)/len(batch[0])
print('Loss: %.5f'%batchloss.item())
optimizer.zero_grad()
batchloss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(),1)
optimizer.step()
for i in range(1,args.EPOCHS+1):
train(i)
if (i>=200)and(i%10 == 0):
torch.save(model.state_dict(),'Saved_Models/TSP_%d/scatgnn_layer_%d_hid_%d_model_%d_temp_%.3f.pth'%(args.num_of_nodes,args.nlayers,args.hidden,i,args.temperature))