我正在尝试使用GAN架构生成Predator-Prey数据集。
显然,两者具有相关性(某些数学函数)。
我需要使用GAN生成它们的数字,而无需找到方程式的参数。
* 我在GAN中相对较新
例如:
Pray = [50. 50. 50. ... 88. 88. 89.] # length 10000
Predator = [100. 99. 99. ... 462. 462. 462.] # length 10000
您如何适应这种情况?
我什至只能使用密集/线性(+ Relu)图层来解决此问题?
我尝试了这一点:(同时从原始数据中采样了1000个值)
class Generator_distribution(nn.Module):
def __init__(self):
# initialize nn Module
super().__init__()
self.layers = nn.ModuleList()
# architecture
self.layers.append(nn.Linear(1000,64))
self.layers.append(nn.LeakyReLU())
self.layers.append(nn.Linear(64,32))
self.layers.append(nn.LeakyReLU())
self.layers.append(nn.Linear(32,1))
def forward(self,input_tensor):
x = input_tensor
for l in self.layers:
x = l(x)
return x
class Discriminator_distribution(nn.Module):
def __init__(self):
super().__init__()
self.layers = nn.ModuleList()
self.layers.append(nn.Linear(1000,32))
self.layers.append(nn.LeakyReLU())
self.layers.append(nn.Linear(32,16))
self.layers.append(nn.LeakyReLU())
self.layers.append(nn.Linear(16,8))
self.layers.append(nn.LeakyReLU())
self.layers.append(nn.Linear(8,1))
self.layers.append(nn.Sigmoid())
def forward(self,input_tensor):
x = input_tensor
for l in self.layers:
print(x.shape)
x = l(x)
return x
# later used : nn.BCELoss() & Adam()