from lib.egate_model import Model
from arguments import args
import argparse
if __name__ == "__main__":
args = args()
device = torch.device(args.device)
N_JOBS = args.N_JOBS
CAP = args.CAP
MAX_COORD = args.MAX_COORD
MAX_DIST = args.MAX_DIST
n_steps=args.N_STEPS
model = Model(4,64,2,16)
model = model.to(device)
envs = create_batch_env()
model.load_state_dict(torch.load('model/v5-500.model'))
states = envs.reset()
print ("before mean cost:",np.mean([env.cost for env in envs.envs]))
_,states,history = roll_out(model,envs,states,n_steps)
print ("after mean cost:",np.mean([env.cost for env in envs.envs]))
history = np.array(history)
a = [env.env.tours() for env in envs.envs]
for i in range(len(a)):
print(i)
print(a[i])
import math
import vrp_env
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch_geometric.data import Data, DataLoader
from lib.utils_train import create_batch_env, train
from lib.egate_model import Model
from arguments import args
import argparse
if __name__ == "__main__":
args = args()
device = torch.device(args.device)
N_JOBS = int(args.N_JOBS)
CAP = int(args.CAP)
batch_size = int(args.BATCH)
MAX_COORD = int(args.MAX_COORD)
MAX_DIST = float(args.MAX_DIST)
LR = float(args.LR)
envs = create_batch_env(128, 99)
model = Model(4, 64, 2, 16)
model = model.to(device)
train(model, envs, 1000, 10, 10, 4)
torch.save(model.state_dict(), "model/v5-latest.model")
import torch.nn as nn
import torch.nn.functional as F
from torch_geometric.data import Data,DataLoader
from lib.utils_train import create_batch_env, train
from lib.egate_model import Model
from arguments import args
import argparse
if __name__ == "__main__":
args = args()
device = torch.device(args.device)
N_JOBS = args.N_JOBS
CAP = args.CAP
MAX_COORD = args.MAX_COORD
MAX_DIST = args.MAX_DIST
LR = args.LR
DEPOT_END = args.DEPOT_END
SERVICE_TIME = args.SERVICE_TIME
TW_WIDTH = args.TW_WIDTH
envs = create_batch_env(128,99)
model = Model(8,64,2,16)
model = model.to(device)
# model.load_state_dict(torch.load("model/v8-tw-iter200-rm25-latest.model"))
train(model,envs,1000,20,10,4,n_remove=10)
torch.save(model.state_dict(), "model/v8-tw-iter200-rm25-latest.model")
device = torch.device(args.device)
N_JOBS = args.N_JOBS
CAP = args.CAP
MAX_COORD = args.MAX_COORD
MAX_DIST = args.MAX_DIST
LR = args.LR
DEPOT_END = args.DEPOT_END
SERVICE_TIME = args.SERVICE_TIME
TW_WIDTH = args.TW_WIDTH
N_ROLLOUT = args.N_ROLLOUT
ROLLOUT_STEPS = args.ROLLOUT_STEPS
N_STEPS = args.N_STEPS
model = Model(8, 64, 2, 16)
model = model.to(device)
model.load_state_dict(torch.load("model/v8-tw-iter200-rm25-latest.model"))
inputs = read_input("data/vrptw_99.npy", N_STEPS)
def eval(batch_size=128, n_steps=100, instance=None):
envs = create_batch_env(batch_size, 99, instance=instance)
states = envs.reset()
states, history, actions, values = roll_out(model, envs, states,
n_steps, False, 64)
best_index = np.argmin([env.best for env in envs.envs])
best_cost = envs.envs[best_index].best
best_tours = envs.envs[best_index].best_sol
print("best cost:", best_cost)
print("best tours:", best_tours)