|
|
import torch
|
|
|
import torch.nn as nn
|
|
|
from NFTModel import ConvLSTMAutoEncoder
|
|
|
import numpy as np
|
|
|
from utils import plotresult_, prepareU
|
|
|
from scipy.io import savemat
|
|
|
|
|
|
mfname='modelnfttrain9.pth'
|
|
|
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
print("Using {} device".format(device))
|
|
|
|
|
|
|
|
|
|
|
|
nt = 2048
|
|
|
nc = 2
|
|
|
|
|
|
model = ConvLSTMAutoEncoder(nc, nc, nt, kernel_size=3, features=[64, 128, 256], lstm_layers=1, forward=False).to(device)
|
|
|
|
|
|
|
|
|
|
|
|
def loss_fn(input, target):
|
|
|
return torch.sqrt(nn.functional.mse_loss(input, target))
|
|
|
|
|
|
print('Load previously saved model: ' + mfname + ' ...')
|
|
|
savedata = torch.load(mfname, map_location=device)
|
|
|
model.load_state_dict(savedata['model'], strict=False)
|
|
|
model.eval()
|
|
|
|
|
|
nt = 2**11
|
|
|
dt = 1.0 / 96e9
|
|
|
T = nt * dt
|
|
|
T0 = np.sqrt(np.abs(-2.172e-26) / 2.0)
|
|
|
TT = T / T0
|
|
|
t = np.arange(nt) / nt * TT - TT / 2.0
|
|
|
A = 0.4
|
|
|
q = A / 250 / np.cosh(t / 250)
|
|
|
x, qlam = plotresult_(model, q)
|
|
|
|
|
|
savedict = {
|
|
|
't' : t,
|
|
|
'x' : x,
|
|
|
'qlam' : qlam
|
|
|
}
|
|
|
|
|
|
savemat("nftinferencesech.mat", savedict)
|
|
|
|
|
|
A = 0.4
|
|
|
q = np.zeros_like(t);
|
|
|
q[abs(t) < 2000] = A / 2000
|
|
|
x, qlam = plotresult_(model, q)
|
|
|
|
|
|
savedict = {
|
|
|
't' : t,
|
|
|
'x' : x,
|
|
|
'qlam' : qlam
|
|
|
}
|
|
|
|
|
|
savemat("nftinferencerect.mat", savedict)
|
|
|
|
|
|
A = 1
|
|
|
q = A / 1500 * np.sinc(t / 1500) * np.exp(1j * np.pi)
|
|
|
x, qlam = plotresult_(model, q)
|
|
|
|
|
|
savedict = {
|
|
|
't' : t,
|
|
|
'x' : x,
|
|
|
'qlam' : qlam
|
|
|
}
|
|
|
|
|
|
savemat("nftinferencesinc.mat", savedict)
|
|
|
|
|
|
|
|
|
|
|
|
|
