171 lines
6.9 KiB
Python
171 lines
6.9 KiB
Python
from __future__ import annotations
|
|
import copy
|
|
import torch
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
import numpy as np
|
|
from torch import optim, Tensor
|
|
|
|
|
|
class Net(nn.Module):
|
|
@staticmethod
|
|
def create_target_weights(input_weight_matrix: Tensor) -> Tensor:
|
|
""" Outputting a tensor with the target weights. """
|
|
|
|
target_weight_matrix = np.arange(len(input_weight_matrix)).reshape(len(input_weight_matrix), 1).astype("f")
|
|
|
|
for i in range(len(input_weight_matrix)):
|
|
target_weight_matrix[i] = input_weight_matrix[i][0]
|
|
|
|
return torch.from_numpy(target_weight_matrix)
|
|
|
|
@staticmethod
|
|
def are_weights_diverged(network_weights):
|
|
""" Testing if the weights are eiter converging to infinity or -infinity. """
|
|
|
|
for layer_id, layer in enumerate(network_weights):
|
|
for cell_id, cell in enumerate(layer):
|
|
for weight_id, weight in enumerate(cell):
|
|
if np.isnan(weight):
|
|
return True
|
|
if np.isinf(weight):
|
|
return True
|
|
return False
|
|
|
|
@staticmethod
|
|
def apply_weights(network: Net, new_weights: Tensor) -> Net:
|
|
""" Changing the weights of a network to new given values. """
|
|
|
|
i = 0
|
|
|
|
for layer_id, layer_name in enumerate(network.state_dict()):
|
|
for line_id, line_values in enumerate(network.state_dict()[layer_name]):
|
|
for weight_id, weight_value in enumerate(network.state_dict()[layer_name][line_id]):
|
|
network.state_dict()[layer_name][line_id][weight_id] = new_weights[i]
|
|
i += 1
|
|
|
|
return network
|
|
|
|
def __init__(self, i_size: int, h_size: int, o_size: int, name=None) -> None:
|
|
super().__init__()
|
|
|
|
self.name = name
|
|
self.input_size = i_size
|
|
self.hidden_size = h_size
|
|
self.out_size = o_size
|
|
|
|
self.no_weights = h_size * (i_size + h_size * (h_size - 1) + o_size)
|
|
|
|
""" Data saved in self.s_train_weights_history & self.s_application_weights_history is used for experiments. """
|
|
self.s_train_weights_history = []
|
|
self.s_application_weights_history = []
|
|
self.loss_history = []
|
|
self.trained = False
|
|
|
|
self.is_fixpoint = ""
|
|
|
|
self.fc1 = nn.Linear(i_size, h_size, False)
|
|
self.fc2 = nn.Linear(h_size, h_size, False)
|
|
self.fc3 = nn.Linear(h_size, o_size, False)
|
|
|
|
def forward(self, x):
|
|
x = self.fc1(x)
|
|
x = self.fc2(x)
|
|
x = self.fc3(x)
|
|
|
|
return x
|
|
|
|
def normalize(self, value):
|
|
""" Normalizing the values >= 1 and adding pow(10, -8) to the values equal to 0 """
|
|
|
|
if value >= 1:
|
|
return value/len(self.state_dict())
|
|
elif value == 0:
|
|
return pow(10, -8)
|
|
else:
|
|
return value
|
|
|
|
def input_weight_matrix(self) -> Tensor:
|
|
""" Calculating the input tensor formed from the weights of the net """
|
|
|
|
# The "4" represents the weightwise coordinates used for the matrix: <value><layer_id><cell_id><positional_id>
|
|
weight_matrix = np.arange(self.no_weights * 4).reshape(self.no_weights, 4).astype("f")
|
|
|
|
i = 0
|
|
|
|
for layer_id, layer_name in enumerate(self.state_dict()):
|
|
for line_id, line_values in enumerate(self.state_dict()[layer_name]):
|
|
for weight_id, weight_value in enumerate(self.state_dict()[layer_name][line_id]):
|
|
weight_matrix[i] = weight_value.item(), self.normalize(layer_id), self.normalize(weight_id), self.normalize(line_id)
|
|
i += 1
|
|
|
|
return torch.from_numpy(weight_matrix)
|
|
|
|
def self_train(self, training_steps: int, log_step_size: int, learning_rate: float, input_data: Tensor, target_data: Tensor) -> (np.ndarray, Tensor, list):
|
|
""" Training a network to predict its own weights in order to self-replicate. """
|
|
|
|
optimizer = optim.SGD(self.parameters(), lr=learning_rate, momentum=0.9)
|
|
self.trained = True
|
|
|
|
for training_step in range(training_steps):
|
|
output = self(input_data)
|
|
loss = F.mse_loss(output, target_data)
|
|
optimizer.zero_grad()
|
|
loss.backward()
|
|
optimizer.step()
|
|
|
|
# Saving the history of the weights after a certain amount of steps (aka log_step_size) for research.
|
|
# If it is a soup/mixed env. save weights only at the end of all training steps (aka a soup/mixed epoch)
|
|
if "soup" not in self.name and "mixed" not in self.name:
|
|
# If self-training steps are lower than 10, then append weight history after each ST step.
|
|
if training_steps < 10:
|
|
self.s_train_weights_history.append(output.T.detach().numpy())
|
|
self.loss_history.append(round(loss.detach().numpy().item(), 5))
|
|
else:
|
|
if training_step % log_step_size == 0:
|
|
self.s_train_weights_history.append(output.T.detach().numpy())
|
|
self.loss_history.append(round(loss.detach().numpy().item(), 5))
|
|
|
|
# Saving weights only at the end of a soup/mixed exp. epoch.
|
|
if "soup" in self.name or "mixed" in self.name:
|
|
self.s_train_weights_history.append(output.T.detach().numpy())
|
|
self.loss_history.append(round(loss.detach().numpy().item(), 5))
|
|
|
|
return output.detach().numpy(), loss, self.loss_history
|
|
|
|
def self_application(self, weights_matrix: Tensor, SA_steps: int, log_step_size: int) -> Net:
|
|
""" Inputting the weights of a network to itself for a number of steps, without backpropagation. """
|
|
|
|
data = copy.deepcopy(weights_matrix)
|
|
new_net = copy.deepcopy(self)
|
|
# output = new_net(data)
|
|
|
|
for i in range(SA_steps):
|
|
output = new_net(data)
|
|
|
|
# Saving the weights history after a certain amount of steps (aka log_step_size) for research purposes.
|
|
# If self-application steps are lower than 10, then append weight history after each SA step.
|
|
if SA_steps < 10:
|
|
self.s_application_weights_history.append(output.T.detach().numpy())
|
|
else:
|
|
if i % log_step_size == 0:
|
|
self.s_application_weights_history.append(output.T.detach().numpy())
|
|
|
|
""" See after how many steps of SA is the output not changing anymore: """
|
|
# print(f"Self-app. step {i+1}: {Experiment.changing_rate(output2, output)}")
|
|
|
|
for j in range(len(data)):
|
|
""" Constructing the weight matrix to have it as the next input. """
|
|
data[j][0] = output[j]
|
|
|
|
new_net = self.apply_weights(new_net, output)
|
|
|
|
return new_net
|
|
|
|
def attack(self, other_net: Net) -> Net:
|
|
other_net_weights = other_net.input_weight_matrix()
|
|
SA_steps = 1
|
|
log_step_size = 1
|
|
|
|
return self.self_application(other_net_weights, SA_steps, log_step_size)
|