File size: 14,265 Bytes
e923d6f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 | import torch
from torch import nn
import numpy as np
import math
from model.temporal_attention import TemporalAttentionLayer
class EmbeddingModule(nn.Module):
def __init__(self, node_features, edge_features, memory, neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features, embedding_dimension, device,
dropout):
super(EmbeddingModule, self).__init__()
self.node_features = node_features
self.edge_features = edge_features
# self.memory = memory
self.neighbor_finder = neighbor_finder
self.time_encoder = time_encoder
self.n_layers = n_layers
self.n_node_features = n_node_features
self.n_edge_features = n_edge_features
self.n_time_features = n_time_features
self.dropout = dropout
self.embedding_dimension = embedding_dimension
self.device = device
def compute_embedding(self, memory, source_nodes, timestamps, n_layers, n_neighbors=20, time_diffs=None,
use_time_proj=True):
return NotImplemented
class IdentityEmbedding(EmbeddingModule):
def compute_embedding(self, memory, source_nodes, timestamps, n_layers, n_neighbors=20, time_diffs=None,
use_time_proj=True):
return memory[source_nodes, :]
class TimeEmbedding(EmbeddingModule):
def __init__(self, node_features, edge_features, memory, neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features, embedding_dimension, device,
n_heads=2, dropout=0.1, use_memory=True, n_neighbors=1):
super(TimeEmbedding, self).__init__(node_features, edge_features, memory,
neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features,
embedding_dimension, device, dropout)
class NormalLinear(nn.Linear):
# From Jodie code
def reset_parameters(self):
stdv = 1. / math.sqrt(self.weight.size(1))
self.weight.data.normal_(0, stdv)
if self.bias is not None:
self.bias.data.normal_(0, stdv)
self.embedding_layer = NormalLinear(1, self.n_node_features)
def compute_embedding(self, memory, source_nodes, timestamps, n_layers, n_neighbors=20, time_diffs=None,
use_time_proj=True):
source_embeddings = memory[source_nodes, :] * (1 + self.embedding_layer(time_diffs.unsqueeze(1)))
return source_embeddings
class GraphEmbedding(EmbeddingModule):
def __init__(self, node_features, edge_features, memory, neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features, embedding_dimension, device,
n_heads=2, dropout=0.1, use_memory=True):
super(GraphEmbedding, self).__init__(node_features, edge_features, memory,
neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features,
embedding_dimension, device, dropout)
self.use_memory = use_memory
self.device = device
def compute_embedding(self, memory, source_nodes, timestamps, n_layers, n_neighbors=20, time_diffs=None,
use_time_proj=True):
"""Recursive implementation of curr_layers temporal graph attention layers.
src_idx_l [batch_size]: users / items input ids.
cut_time_l [batch_size]: scalar representing the instant of the time where we want to extract the user / item representation.
curr_layers [scalar]: number of temporal convolutional layers to stack.
num_neighbors [scalar]: number of temporal neighbor to consider in each convolutional layer.
"""
assert (n_layers >= 0)
source_nodes_torch = torch.from_numpy(source_nodes).long().to(self.device)
timestamps_torch = torch.unsqueeze(torch.from_numpy(timestamps).float().to(self.device), dim=1)
# query node always has the start time -> time span == 0
source_nodes_time_embedding = self.time_encoder(torch.zeros_like(
timestamps_torch))
source_node_features = self.node_features[source_nodes_torch, :]
if self.use_memory:
source_node_features = memory[source_nodes, :] + source_node_features
if n_layers == 0:
return source_node_features
else:
source_node_conv_embeddings = self.compute_embedding(memory,
source_nodes,
timestamps,
n_layers=n_layers - 1,
n_neighbors=n_neighbors)
neighbors, edge_idxs, edge_times = self.neighbor_finder.get_temporal_neighbor(
source_nodes,
timestamps,
n_neighbors=n_neighbors)
neighbors_torch = torch.from_numpy(neighbors).long().to(self.device)
edge_idxs = torch.from_numpy(edge_idxs).long().to(self.device)
edge_deltas = timestamps[:, np.newaxis] - edge_times
edge_deltas_torch = torch.from_numpy(edge_deltas).float().to(self.device)
neighbors = neighbors.flatten()
neighbor_embeddings = self.compute_embedding(memory,
neighbors,
np.repeat(timestamps, n_neighbors),
n_layers=n_layers - 1,
n_neighbors=n_neighbors)
effective_n_neighbors = n_neighbors if n_neighbors > 0 else 1
neighbor_embeddings = neighbor_embeddings.view(len(source_nodes), effective_n_neighbors, -1)
edge_time_embeddings = self.time_encoder(edge_deltas_torch)
edge_features = self.edge_features[edge_idxs, :]
mask = neighbors_torch == 0
source_embedding = self.aggregate(n_layers, source_node_conv_embeddings,
source_nodes_time_embedding,
neighbor_embeddings,
edge_time_embeddings,
edge_features,
mask)
return source_embedding
def aggregate(self, n_layers, source_node_features, source_nodes_time_embedding,
neighbor_embeddings,
edge_time_embeddings, edge_features, mask):
return NotImplemented
class GraphSumEmbedding(GraphEmbedding):
def __init__(self, node_features, edge_features, memory, neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features, embedding_dimension, device,
n_heads=2, dropout=0.1, use_memory=True):
super(GraphSumEmbedding, self).__init__(node_features=node_features,
edge_features=edge_features,
memory=memory,
neighbor_finder=neighbor_finder,
time_encoder=time_encoder, n_layers=n_layers,
n_node_features=n_node_features,
n_edge_features=n_edge_features,
n_time_features=n_time_features,
embedding_dimension=embedding_dimension,
device=device,
n_heads=n_heads, dropout=dropout,
use_memory=use_memory)
self.linear_1 = torch.nn.ModuleList([torch.nn.Linear(embedding_dimension + n_time_features +
n_edge_features, embedding_dimension)
for _ in range(n_layers)])
self.linear_2 = torch.nn.ModuleList(
[torch.nn.Linear(embedding_dimension + n_node_features + n_time_features,
embedding_dimension) for _ in range(n_layers)])
def aggregate(self, n_layer, source_node_features, source_nodes_time_embedding,
neighbor_embeddings,
edge_time_embeddings, edge_features, mask):
neighbors_features = torch.cat([neighbor_embeddings, edge_time_embeddings, edge_features],
dim=2)
neighbor_embeddings = self.linear_1[n_layer - 1](neighbors_features)
neighbors_sum = torch.nn.functional.relu(torch.sum(neighbor_embeddings, dim=1))
source_features = torch.cat([source_node_features,
source_nodes_time_embedding.squeeze()], dim=1)
source_embedding = torch.cat([neighbors_sum, source_features], dim=1)
source_embedding = self.linear_2[n_layer - 1](source_embedding)
return source_embedding
class GraphAttentionEmbedding(GraphEmbedding):
def __init__(self, node_features, edge_features, memory, neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features, n_time_features, embedding_dimension, device,
n_heads=2, dropout=0.1, use_memory=True):
super(GraphAttentionEmbedding, self).__init__(node_features, edge_features, memory,
neighbor_finder, time_encoder, n_layers,
n_node_features, n_edge_features,
n_time_features,
embedding_dimension, device,
n_heads, dropout,
use_memory)
self.attention_models = torch.nn.ModuleList([TemporalAttentionLayer(
n_node_features=n_node_features,
n_neighbors_features=n_node_features,
n_edge_features=n_edge_features,
time_dim=n_time_features,
n_head=n_heads,
dropout=dropout,
output_dimension=n_node_features)
for _ in range(n_layers)])
def aggregate(self, n_layer, source_node_features, source_nodes_time_embedding,
neighbor_embeddings,
edge_time_embeddings, edge_features, mask):
attention_model = self.attention_models[n_layer - 1]
source_embedding, _ = attention_model(source_node_features,
source_nodes_time_embedding,
neighbor_embeddings,
edge_time_embeddings,
edge_features,
mask)
return source_embedding
def get_embedding_module(module_type, node_features, edge_features, memory, neighbor_finder,
time_encoder, n_layers, n_node_features, n_edge_features, n_time_features,
embedding_dimension, device,
n_heads=2, dropout=0.1, n_neighbors=None,
use_memory=True):
if module_type == "graph_attention":
return GraphAttentionEmbedding(node_features=node_features,
edge_features=edge_features,
memory=memory,
neighbor_finder=neighbor_finder,
time_encoder=time_encoder,
n_layers=n_layers,
n_node_features=n_node_features,
n_edge_features=n_edge_features,
n_time_features=n_time_features,
embedding_dimension=embedding_dimension,
device=device,
n_heads=n_heads, dropout=dropout, use_memory=use_memory)
elif module_type == "graph_sum":
return GraphSumEmbedding(node_features=node_features,
edge_features=edge_features,
memory=memory,
neighbor_finder=neighbor_finder,
time_encoder=time_encoder,
n_layers=n_layers,
n_node_features=n_node_features,
n_edge_features=n_edge_features,
n_time_features=n_time_features,
embedding_dimension=embedding_dimension,
device=device,
n_heads=n_heads, dropout=dropout, use_memory=use_memory)
elif module_type == "identity":
return IdentityEmbedding(node_features=node_features,
edge_features=edge_features,
memory=memory,
neighbor_finder=neighbor_finder,
time_encoder=time_encoder,
n_layers=n_layers,
n_node_features=n_node_features,
n_edge_features=n_edge_features,
n_time_features=n_time_features,
embedding_dimension=embedding_dimension,
device=device,
dropout=dropout)
elif module_type == "time":
return TimeEmbedding(node_features=node_features,
edge_features=edge_features,
memory=memory,
neighbor_finder=neighbor_finder,
time_encoder=time_encoder,
n_layers=n_layers,
n_node_features=n_node_features,
n_edge_features=n_edge_features,
n_time_features=n_time_features,
embedding_dimension=embedding_dimension,
device=device,
dropout=dropout,
n_neighbors=n_neighbors)
else:
raise ValueError("Embedding Module {} not supported".format(module_type))
|