czzhangheng
/
FS-TFP
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

more clear

This commit is contained in:
HengZhang 2024-11-27 13:52:14 +08:00
parent d3021c8112
commit b335047b87
5 changed files with 71 additions and 76 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
federatedscope/core/configs/cfg_model.py
View File

@ -49,20 +49,6 @@ def extend_model_cfg(cfg):
cfg.model.contrast_topk = 100
cfg.model.contrast_temp = 1.0
# Traffic Flow model parameters, These are only default values.
# Please modify the specific parameters directly in the baselines/YAML files.
cfg.model.num_nodes = 0
cfg.model.rnn_units = 64
cfg.model.dropout = 0.1
cfg.model.horizon = 12
cfg.model.input_dim = 1 # If 0, model will be built by data.shape
cfg.model.output_dim = 1
cfg.model.embed_dim = 10
cfg.model.num_layers = 1 # In GPR-GNN, K = layer
cfg.model.cheb_order = 1 # A tuple, e.g., (in_channel, h, w)
cfg.model.use_day = True
cfg.model.use_week = True
# ---------------------------------------------------------------------- #
# Criterion related options

37
federatedscope/core/configs/cfg_trafficflow.py
View File

@ -12,24 +12,29 @@ def extend_trafficflow_cfg(cfg):
# ---------------------------------------------------------------------- #
# Model related options
# ---------------------------------------------------------------------- #
cfg.model = CN()
cfg.model.tfp = CN()
cfg.model.model_num_per_trainer = 1 # some methods may leverage more
cfg.model.tfp.model_num_per_trainer = 1 # some methods may leverage more
# than one model in each trainer
cfg.model.type = 'trafficflow'
cfg.model.use_bias = True
cfg.model.task = 'trafficflowprediction'
cfg.model.num_nodes = 0
cfg.model.rnn_units = 64
cfg.model.dropout = 0.1
cfg.model.horizon = 12
cfg.model.input_dim = 1 # If 0, model will be built by data.shape
cfg.model.output_dim = 1
cfg.model.embed_dim = 10
cfg.model.num_layers = 1 # In GPR-GNN, K = layer
cfg.model.cheb_order = 1 # A tuple, e.g., (in_channel, h, w)
cfg.model.use_day = True
cfg.model.use_week = True
# cfg.tfp.model.type = 'trafficflow'
# cfg.tfp.model.use_bias = True
# cfg.tfp.model.task = 'trafficflowprediction'
cfg.model.tfp.num_nodes = 0
cfg.model.tfp.rnn_units = 64
cfg.model.tfp.dropout = 0.1
cfg.model.tfp.horizon = 12
cfg.model.tfp.input_dim = 1 # If 0, model will be built by data.shape
cfg.model.tfp.output_dim = 1
cfg.model.tfp.embed_dim = 10
cfg.model.tfp.num_layers = 1 # In GPR-GNN, K = layer
cfg.model.tfp.cheb_order = 1 # A tuple, e.g., (in_channel, h, w)
cfg.model.tfp.use_day = True
cfg.model.tfp.use_week = True
cfg.model.tfp.minigraph = CN()
cfg.model.tfp.minigraph.enable = False
cfg.model.tfp.minigraph.size = 5
# ---------------------------------------------------------------------- #
# Criterion related options

20
federatedscope/trafficflow/dataloader/traffic_dataloader.py
View File

@ -125,22 +125,22 @@ def load_traffic_data(config, client_cfgs):
# normalize st data
normalizer = 'std'
scaler = normalize_dataset(x_train[..., :config.model.input_dim], normalizer, config.data.column_wise)
scaler = normalize_dataset(x_train[..., :config.model.tfp.input_dim], normalizer, config.data.column_wise)
config.data.scaler = [float(scaler.mean), float(scaler.std)]
x_train[..., :config.model.input_dim] = scaler.transform(x_train[..., :config.model.input_dim])
x_val[..., :config.model.input_dim] = scaler.transform(x_val[..., :config.model.input_dim])
x_test[..., :config.model.input_dim] = scaler.transform(x_test[..., :config.model.input_dim])
# y_train[..., :config.model.output_dim] = scaler.transform(y_train[..., :config.model.output_dim])
# y_val[..., :config.model.output_dim] = scaler.transform(y_val[..., :config.model.output_dim])
# y_test[..., :config.model.output_dim] = scaler.transform(y_test[..., :config.model.output_dim])
x_train[..., :config.model.tfp.input_dim] = scaler.transform(x_train[..., :config.model.tfp.input_dim])
x_val[..., :config.model.tfp.input_dim] = scaler.transform(x_val[..., :config.model.tfp.input_dim])
x_test[..., :config.model.tfp.input_dim] = scaler.transform(x_test[..., :config.model.tfp.input_dim])
# y_train[..., :config.model.tfp.output_dim] = scaler.transform(y_train[..., :config.model.tfp.output_dim])
# y_val[..., :config.model.tfp.output_dim] = scaler.transform(y_val[..., :config.model.tfp.output_dim])
# y_test[..., :config.model.tfp.output_dim] = scaler.transform(y_test[..., :config.model.tfp.output_dim])
# Client-side dataset splitting
node_num = config.data.num_nodes
client_num = config.federate.client_num
per_samples = node_num // client_num
data_list, cur_index = dict(), 0
input_dim, output_dim = config.model.input_dim, config.model.output_dim
input_dim, output_dim = config.model.tfp.input_dim, config.model.tfp.output_dim
for i in range(client_num):
if cur_index + per_samples <= node_num:
# Normal slicing
@ -156,7 +156,7 @@ def load_traffic_data(config, client_cfgs):
sub_array_train = x_train[:, :, cur_index:cur_index + per_samples, :]
sub_array_val = x_val[:, :, cur_index:cur_index + per_samples, :]
sub_array_test = x_test[:, :, cur_index:cur_index + per_samples, :]
padding = np.zeros((x_train.shape[0], config.data.lag ,config.data.lag, per_samples - x_train.shape[1], config.model.output_dim))
padding = np.zeros((x_train.shape[0], config.data.lag ,config.data.lag, per_samples - x_train.shape[1], config.model.tfp.output_dim))
sub_array_train = np.concatenate((sub_array_train, padding), axis=2)
sub_array_val = np.concatenate((sub_array_val, padding), axis=2)
sub_array_test = np.concatenate((sub_array_test, padding), axis=2)
@ -185,7 +185,7 @@ def load_traffic_data(config, client_cfgs):
)
}
cur_index += per_samples
config.model.num_nodes = per_samples
config.model.tfp.num_nodes = per_samples
return data_list, config

48
federatedscope/trafficflow/model/FedDGCN.py
View File

@ -40,33 +40,33 @@ class DGCRM(nn.Module):
class FedDGCN(nn.Module):
def __init__(self, args):
super(FedDGCN, self).__init__()
self.num_node = args.num_nodes
self.input_dim = args.input_dim
self.hidden_dim = args.rnn_units
self.output_dim = args.output_dim
self.horizon = args.horizon
self.num_layers = args.num_layers
self.use_D = args.use_day
self.use_W = args.use_week
self.dropout1 = nn.Dropout(p=args.dropout) # 0.1
self.dropout2 = nn.Dropout(p=args.dropout)
self.node_embeddings1 = nn.Parameter(torch.randn(self.num_node, args.embed_dim), requires_grad=True)
self.node_embeddings2 = nn.Parameter(torch.randn(self.num_node, args.embed_dim), requires_grad=True)
self.T_i_D_emb = nn.Parameter(torch.empty(288, args.embed_dim))
self.D_i_W_emb = nn.Parameter(torch.empty(7, args.embed_dim))
self.num_node = args.tfp.num_nodes
self.input_dim = args.tfp.input_dim
self.hidden_dim = args.tfp.rnn_units
self.output_dim = args.tfp.output_dim
self.horizon = args.tfp.horizon
self.num_layers = args.tfp.num_layers
self.use_D = args.tfp.use_day
self.use_W = args.tfp.use_week
self.dropout1 = nn.Dropout(p=args.tfp.dropout) # 0.1
self.dropout2 = nn.Dropout(p=args.tfp.dropout)
self.node_embeddings1 = nn.Parameter(torch.randn(self.num_node, args.tfp.embed_dim), requires_grad=True)
self.node_embeddings2 = nn.Parameter(torch.randn(self.num_node, args.tfp.embed_dim), requires_grad=True)
self.T_i_D_emb = nn.Parameter(torch.empty(288, args.tfp.embed_dim))
self.D_i_W_emb = nn.Parameter(torch.empty(7, args.tfp.embed_dim))
# Initialize parameters
nn.init.xavier_uniform_(self.node_embeddings1)
nn.init.xavier_uniform_(self.T_i_D_emb)
nn.init.xavier_uniform_(self.D_i_W_emb)
self.encoder1 = DGCRM(args.num_nodes, args.input_dim, args.rnn_units, args.cheb_order,
args.embed_dim, args.num_layers)
self.encoder2 = DGCRM(args.num_nodes, args.input_dim, args.rnn_units, args.cheb_order,
args.embed_dim, args.num_layers)
self.encoder1 = DGCRM(args.tfp.num_nodes, args.tfp.input_dim, args.tfp.rnn_units, args.tfp.cheb_order,
args.tfp.embed_dim, args.tfp.num_layers)
self.encoder2 = DGCRM(args.tfp.num_nodes, args.tfp.input_dim, args.tfp.rnn_units, args.tfp.cheb_order,
args.tfp.embed_dim, args.tfp.num_layers)
# predictor
self.end_conv1 = nn.Conv2d(1, args.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
self.end_conv2 = nn.Conv2d(1, args.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
self.end_conv3 = nn.Conv2d(1, args.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
self.end_conv1 = nn.Conv2d(1, args.tfp.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
self.end_conv2 = nn.Conv2d(1, args.tfp.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
self.end_conv3 = nn.Conv2d(1, args.tfp.horizon * self.output_dim, kernel_size=(1, self.hidden_dim), bias=True)
def forward(self, source, i=2):
node_embedding1 = self.node_embeddings1
@ -107,10 +107,10 @@ def ModelBuilder(model_config, local_data):
return model
def call_ddgcrn(model_config, local_data):
if model_config.type == "DDGCRN":
def call_feddgcn(model_config, local_data):
if model_config.type == "FedDGCN":
model = ModelBuilder(model_config, local_data)
return model
register_model("DDGCRN", call_ddgcrn)
register_model("FedDGCN", call_feddgcn)

28
scripts/trafficflow_exp_scripts/D4.yaml
View File

@ -1,4 +1,4 @@
use_gpu: True
use_gpu: False
seed: 10
device: 0
early_stop:
@ -33,17 +33,21 @@ dataloader:
model:
type: FedDGCN
task: TrafficFlowPrediction
dropout: 0.1
horizon: 12
num_nodes: 0
input_dim: 1
output_dim: 1
embed_dim: 10
rnn_units: 64
num_layers: 1
cheb_order: 2
use_day: True
use_week: True
tfp:
dropout: 0.1
horizon: 12
num_nodes: 0
input_dim: 1
output_dim: 1
embed_dim: 10
rnn_units: 64
num_layers: 1
cheb_order: 2
use_day: True
use_week: True
minigraph:
enable: True
size: 5
train:
batch_or_epoch: 'epoch'
local_update_steps: 1