forked from LeoVogiatzis/GNN_based_NILM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Auto_Encoder.py
253 lines (204 loc) · 9.02 KB
/
Auto_Encoder.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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
import sklearn
import torch
from torch.utils.data import TensorDataset, DataLoader
from torch_geometric.utils import negative_sampling, remove_self_loops, add_self_loops
from torch_geometric.nn.inits import reset
from torch_geometric.nn import GCNConv, BatchNorm
import pandas as pd
import torch
import torch_geometric
from torch_geometric.data import Dataset, Data
import numpy as np
import os
from tqdm import tqdm
print(f"Torch version: {torch.__version__}")
print(f"Cuda available: {torch.cuda.is_available()}")
print(f"Torch geometric version: {torch_geometric.__version__}")
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from torch_geometric.transforms import RandomLinkSplit, RandomNodeSplit
import torch
from torch.nn import Linear
from torch_geometric.nn import GCNConv
import torch.nn.functional as F
import math
from sklearn.model_selection import train_test_split
EPS = 1e-15
MAX_LOGSTD = 10
epochs = 20
class GCNEncoder(torch.nn.Module):
def __init__(self, dims, dropout):
super(GCNEncoder, self).__init__()
self.dropout = dropout
self.layers = torch.nn.ModuleList()
for i in range(len(dims) - 1):
conv = GCNConv(dims[i], dims[i + 1])
self.layers.append(conv)
def forward(self, x, edge_index):
num_layers = len(self.layers)
for idx, layer in enumerate(self.layers):
if idx < num_layers - 1:
x = layer(x, edge_index)
x = F.relu(x)
x = F.dropout(x, p=self.dropout, training=self.training)
else:
x = layer(x, edge_index)
return x
class InnerProductDecoder(torch.nn.Module):
def forward(self, z, edge_index, sigmoid=True):
value = (z[edge_index[0]] * z[edge_index[1]]).sum(dim=1)
return torch.sigmoid(value) if sigmoid else value
class GAE(torch.nn.Module):
def __init__(self, dims, dropout):
super(GAE, self).__init__()
self.encoder = GCNEncoder(dims, dropout)
self.decoder = InnerProductDecoder()
GAE.reset_parameters(self)
def reset_parameters(self):
reset(self.encoder)
reset(self.decoder)
def encode(self, *args, **kwargs):
return self.encoder(*args, **kwargs)
def decode(self, *args, **kwargs):
return self.decoder(*args, **kwargs)
def recon_loss(self, z, pos_edge_index, neg_edge_index=None):
"""Given latent variables, computes the binary cross entropy loss for positive edges and negative sampled edges.
Args:
z (Tensor): The latent space representations.
pos_edge_index (LongTensor): The positive edges to train against.
neg_edge_index (LongTensor, optional): The negative edges to train against.
If not given, uses negative sampling to calculate negative edges.
"""
pos_loss = -torch.log(self.decoder(z, pos_edge_index, sigmoid=True) + EPS).mean()
# Do not include self-loops in negative samples
pos_edge_index, _ = remove_self_loops(pos_edge_index)
pos_edge_index, _ = add_self_loops(pos_edge_index)
if neg_edge_index is None:
neg_edge_index = negative_sampling(pos_edge_index, z.size(0))
neg_loss = -torch.log(1 - self.decoder(z, neg_edge_index, sigmoid=True) + EPS).mean()
return pos_loss + neg_loss
class NilmDataset(Dataset):
def __init__(self, root, filename, window, sigma, test=False, transform=None, pre_transform=None):
"""2
root = Where the dataset should be stored. This folder is split
into raw_dir (downloaded dataset) and processed_dir (processed data).
"""
self.test = test
self.filename = filename
self.window = window
self.sigma = sigma
super(NilmDataset, self).__init__(root, transform, pre_transform)
@property
def raw_file_names(self):
""" If this file exists in raw_dir, the download is not triggered.
(The download func. is not implemented here)
"""
return [self.filename]
@property
def processed_file_names(self):
""" If these files are found in raw_dir, processing is skipped"""
data = pd.read_csv(self.raw_paths[0]).reset_index()
if self.test:
return [f'data_test_{i}.pt' for i in list(data.index)]
else:
return [f'data_{i}.pt' for i in list(data.index)]
def download(self):
pass
def process(self):
idx = 0
for raw_path in self.raw_paths:
appliance = pd.read_csv(raw_path).reset_index()
main_val = appliance['dishwaser_20'].values # get only readings
data_vec = main_val
adjacency, drift = self._get_adjacency_info(data_vec)
edge_indices = self._to_edge_index(adjacency)
# node_feats = np.asarray(drift)
# node_feats = node_feats.reshape((-1, 1))
# node_feats = torch.tensor(node_feats, dtype=torch.float)
labels = np.asarray(drift)
labels = torch.tensor(labels, dtype=torch.int64)
data = Data(edge_index=edge_indices, y=labels,
# train_mask=[2000], test_mask=[2000]
)
if self.pre_filter is not None and not self.pre_filter(data):
continue
if self.pre_transform is not None:
data = self.pre_transform(data)
if self.test:
torch.save(data, os.path.join(self.processed_dir, f'data_test_{idx}.pt'))
else:
torch.save(data, os.path.join(self.processed_dir, f'data_{idx}.pt'))
idx += 1
def _get_node_features(self, graph):
"""
This will return a matrix / 2d array of the shape
[Number of Nodes, Node Feature size]
We could also use torch_geometric.from_networkx to create a Data object
with both adjacency and features, but instead we do it manually here
"""
all_node_feats = list(nx.get_node_attributes(graph, 'drift').values())
all_node_feats = np.asarray(all_node_feats)
all_node_feats = all_node_feats.reshape((-1, 1))
return torch.tensor(all_node_feats, dtype=torch.float)
def _get_adjacency_info(self, data_vec):
data_aggr = []
for k in range(0, int(np.floor(len(data_vec) / self.window))):
data_aggr.append(np.mean(data_vec[k * self.window:((k + 1) * self.window)]))
if (len(data_vec) % self.window > 0):
data_aggr.append(np.mean(data_vec[int(np.floor(len(data_vec) / self.window)) * self.window:]))
delta_p = [np.round(data_aggr[i + 1] - data_aggr[i], 2) for i in range(0, len(data_aggr) - 1)]
Am = np.zeros((len(delta_p), len(delta_p)))
for i in range(0, Am.shape[0]):
for j in range(0, Am.shape[1]):
Am[i, j] = math.exp(-((delta_p[i] - delta_p[j]) / self.sigma) ** 2)
Am = np.where(Am >= 0.5, 1, 0)
return Am, delta_p
def _to_edge_index(self, adjacency):
edge_indices = []
for i in range(0, adjacency.shape[0]):
for j in range(i, adjacency.shape[0]):
if adjacency[i, j] != 0.0:
edge_indices += [[i, j], [j, i]]
edge_indices = torch.tensor(edge_indices)
edge_indices = edge_indices.t().to(torch.long).view(2, -1)
return edge_indices
def len(self):
return len(self.processed_file_names)
def get(self, idx):
""" - Equivalent to __getitem__ in pytorch
- Is not needed for PyG's InMemoryDataset
"""
if self.test:
data = torch.load(os.path.join(self.processed_dir, f'data_test_{idx}.pt'))
else:
data = torch.load(os.path.join(self.processed_dir, f'data_{idx}.pt'))
return data
def main():
from torch_geometric.datasets import Planetoid
from torch_geometric.transforms import NormalizeFeatures
dataset = Planetoid(root='data/Planetoid', name='Cora', transform=NormalizeFeatures())
data = dataset[0]
print(data.y)
# ------------------------------------------------------------------------------------
# dataset = NilmDataset(root='data', filename='dishwasher.csv', window=20, sigma=20)
# data = dataset[0]
# degrees = torch_geometric.utils.degree(index=data.edge_index[0])
# data.x = degrees
# print(data)
# ------------------------------------------------------------------------------------
model = GAE([2, 2], dropout=0.2)
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
criterion = torch.nn.MSELoss()
for i in range(epochs):
optimizer.zero_grad()
# z = model.encode(data.x, data.train_pos_edge_index)
z = model.encode(data.x, data.edge_index)
out = model.decode(z, data.edge_index)
loss = criterion(data.x, out, sigmoid=True)
# loss = model.recon_loss(z, data.train_pos_edge_index)
loss.backward()
optimizer.step()
model.encode(data.x, data.edge_index)
print(model)
if __name__ == '__main__':
main()