Skip to content

B
BTS-MTGNN
Commit e0711e14 by zlj
Options

add time count

parent ae8b45b0
Showing with 658 additions and 7 deletions
examples/train_static.py 0 → 100644
import argparse
import os
import profile
import sys
import psutil
from os.path import abspath, join, dirname
current_path = os.path.dirname(os.path.abspath(__file__))
parent_path = os.path.abspath(os.path.join(current_path, os.pardir))
sys.path.append(parent_path)
from starrygl.sample.part_utils.transformer_from_speed import load_from_shared_node_partition, load_from_speed
from starrygl.sample.count_static import time_count
from starrygl.sample.sample_core.LocalNegSampling import LocalNegativeSampling
from starrygl.distributed.context import DistributedContext
from starrygl.distributed.utils import DistIndex
from starrygl.module.modules import GeneralModel
from pathlib import Path
from pathlib import Path
from starrygl.module.utils import parse_config
from starrygl.sample.cache.fetch_cache import FetchFeatureCache
from starrygl.sample.graph_core import DataSet, DistributedGraphStore, TemporalNeighborSampleGraph
from starrygl.module.utils import parse_config, EarlyStopMonitor
from starrygl.sample.graph_core import DataSet, DistributedGraphStore, TemporalNeighborSampleGraph
from starrygl.sample.memory.shared_mailbox import SharedMailBox
from starrygl.sample.sample_core.base import NegativeSampling
from starrygl.sample.sample_core.neighbor_sampler import NeighborSampler
from starrygl.sample.part_utils.partition_tgnn import partition_load
import torch
import time
import torch
import torch.nn.functional as F
import torch.distributed as dist
import torch.multiprocessing as mp
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.distributed import init_process_group, destroy_process_group
from starrygl.sample.count_static import time_count as tt
import os
from starrygl.sample.data_loader import DistributedDataLoader
from starrygl.sample.batch_data import SAMPLE_TYPE
from starrygl.sample.stream_manager import getPipelineManger
from torch.profiler import profile, record_function, ProfilerActivity
parser = argparse.ArgumentParser(
description="RPC Reinforcement Learning Example",
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
)
parser.add_argument('--rank', default=0, type=int, metavar='W',
help='name of dataset')
parser.add_argument('--local_rank', default=0, type=int, metavar='W',
help='name of dataset')
parser.add_argument('--patience', type=int, default=20, help='Patience for early stopping')
parser.add_argument('--world_size', default=1, type=int, metavar='W',
help='number of negative samples')
parser.add_argument('--dataname', default="WIKI", type=str, metavar='W',
help='name of dataset')
parser.add_argument('--model', default='TGN', type=str, metavar='W',
help='name of model')
parser.add_argument('--part_test', default='part', type=str, metavar='W',
help='name of model')
parser.add_argument('--partition', default='part', type=str, metavar='W',
help='name of model')
parser.add_argument('--topk', default='0', type=str, metavar='W',
help='name of model')
parser.add_argument('--probability', default=0.1, type=float, metavar='W',
help='name of model')
parser.add_argument('--sample_type', default='recent', type=str, metavar='W',
help='name of model')
parser.add_argument('--local_neg_sample', default=True, type=bool, metavar='W',
help='name of model')
parser.add_argument('--shared_memory_ssim', default=2, type=float, metavar='W',
help='name of model')
parser.add_argument('--neg_samples', default=1, type=int, metavar='W',
help='name of model')
parser.add_argument('--eval_neg_samples', default=1, type=int, metavar='W',
help='name of model')
parser.add_argument('--memory_type', default='all_update', type=str, metavar='W',
help='name of model')
#boundery_recent_uniform boundery_recent_decay
args = parser.parse_args()
if args.memory_type == 'all_local' or args.topk != '0':
train_cross_probability = 0
else:
train_cross_probability = 1
if args.memory_type == 'all_local':
args.sample_type = 'boundery_recent_uniform'
args.probability = 0
from sklearn.metrics import average_precision_score, roc_auc_score
import torch
import time
import random
import numpy as np
from sklearn.metrics import average_precision_score, roc_auc_score
from torch.nn.parallel import DistributedDataParallel as DDP
def get_network_interfaces_with_prefix(prefixs):
interfaces = psutil.net_if_addrs()
matching_interfaces = [iface for iface in interfaces if iface.startswith(prefixs[0]) or iface.startswith(prefixs[1])]
return matching_interfaces
# Example usage
prefix = ("ens4f1np1","ens6f0np0")
matching_interfaces = get_network_interfaces_with_prefix(prefix)
print(f"Network interfaces with prefix '{prefix}': {matching_interfaces}")
#os.environ['CUDA_VISIBLE_DEVICES'] = '2'#str(args.rank)
if not 'WORLD_SIZE' in os.environ:
os.environ["RANK"] = str(args.rank)
os.environ["WORLD_SIZE"] = str(args.world_size)
os.environ["LOCAL_RANK"] = str(args.local_rank)
if not 'MASTER_ADDR' in os.environ:
os.environ["MASTER_ADDR"] = '192.168.2.107'
if not 'MASTER_PORT' in os.environ:
os.environ["MASTER_PORT"] = '9337'
os.environ["NCCL_IB_DISABLE"]='1'
os.environ['NCCL_SOCKET_IFNAME']=matching_interfaces[0]
print('rank {}'.format(int(os.environ["LOCAL_RANK"])))
torch.cuda.set_device(int(os.environ["LOCAL_RANK"]))
local_rank = int(os.environ["LOCAL_RANK"])
def seed_everything(seed=42):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
total_next_batch = 0
total_forward = 0
total_count_score = 0
total_backward = 0
total_prepare_mail = 0
total_update_mail = 0
total_update_memory =0
total_remote_update = 0
def count_empty():
global total_next_batch
global total_forward
global total_count_score
global total_backward
global total_prepare_mail
global total_update_mail
global total_update_memory
global total_remote_update
total_next_batch = 0
total_forward = 0
total_count_score = 0
total_backward = 0
total_prepare_mail = 0
total_update_mail = 0
total_update_memory =0
total_remote_update = 0
def add(t1,t2,t3,t4,t5,t6,t7,t8):
global total_next_batch
global total_forward
global total_count_score
global total_backward
global total_prepare_mail
global total_update_mail
global total_update_memory
global total_remote_update
total_next_batch += t1
total_forward += t2
total_count_score += t4
total_backward += t3
total_prepare_mail += t5
total_update_mail += t6
total_update_memory +=t7
total_remote_update += t8
def query():
global total_next_batch
global total_forward
global total_count_score
global total_backward
global total_prepare_mail
global total_update_mail
global total_update_memory
global total_remote_update
global total_next_batch
return {
"total_next_batch":total_next_batch,
"total_forward" :total_forward ,
"total_count_score" :total_count_score ,
"total_backward" :total_backward ,
"total_prepare_mail" :total_prepare_mail ,
"total_update_mail" :total_update_mail ,
"total_update_memory":total_update_memory,
"total_remote_update":total_remote_update,}
seed_everything(34)
def main():
#torch.autograd.set_detect_anomaly(True)
print('LOCAL RANK {}, RANK{}'.format(os.environ["LOCAL_RANK"],os.environ["RANK"]))
use_cuda = True
sample_param, memory_param, gnn_param, train_param = parse_config('../config/{}.yml'.format(args.model))
memory_param['mode'] = args.memory_type
ctx = DistributedContext.init(backend="nccl", use_gpu=True,memory_group_num=1,cache_use_rpc=True)
torch.set_num_threads(10)
device_id = torch.cuda.current_device()
graph,full_sampler_graph,train_mask,val_mask,test_mask,full_train_mask,cache_route = load_from_speed(args.dataname,seed=123457,top=args.topk,sampler_graph_add_rev=True, feature_device=torch.device('cuda:{}'.format(ctx.local_rank)),partition=args.partition)#torch.device('cpu'))
# 确保 CUDA 可用
if torch.cuda.is_available():
print("Total GPU memory: ", torch.cuda.get_device_properties(0).total_memory/1024**3)
print("Current GPU memory allocated: ", torch.cuda.memory_allocated(0)/1024**3)
print("Current GPU memory reserved: ", torch.cuda.memory_reserved(0)/1024**3)
print("Max GPU memory allocated during this session: ", torch.cuda.max_memory_allocated(0))
print("Max GPU memory reserved during this session: ", torch.cuda.max_memory_reserved(0))
else:
print("CUDA is not available.")
full_dst = full_sampler_graph['edge_index'][1,torch.arange(0,full_sampler_graph['edge_index'].shape[1],2)]
sample_graph = TemporalNeighborSampleGraph(full_sampler_graph,mode = 'full',dist_eid_mapper=graph.eids_mapper)#,local_eids=graph.eids)
eval_sample_graph = TemporalNeighborSampleGraph(full_sampler_graph,mode = 'full',dist_eid_mapper=graph.eids_mapper)
Path("../saved_models/").mkdir(parents=True, exist_ok=True)
Path("../saved_checkpoints/").mkdir(parents=True, exist_ok=True)
get_checkpoint_path = lambda \
epoch: f'../saved_checkpoints/{args.model}-{args.dataname}-{epoch}.pth'
gnn_param['dyrep'] = True if args.model == 'DyRep' else False
use_src_emb = gnn_param['use_src_emb'] if 'use_src_emb' in gnn_param else False
use_dst_emb = gnn_param['use_dst_emb'] if 'use_dst_emb' in gnn_param else False
fanout = []
num_layers = sample_param['layer'] if 'layer' in sample_param else 1
fanout = sample_param['neighbor'] if 'neighbor' in sample_param else [10]
policy = sample_param['strategy'] if 'strategy' in sample_param else 'recent'
policy_train = args.sample_type#'boundery_recent_decay'
if memory_param['type'] != 'none':
mailbox = SharedMailBox(graph.ids.shape[0], memory_param, dim_edge_feat = graph.efeat.shape[1] if graph.efeat is not None else 0,
shared_nodes_index=graph.shared_nids_list[ctx.memory_group_rank],device = torch.device('cuda:{}'.format(local_rank)),cache_route = cache_route,shared_ssim=args.shared_memory_ssim)
else:
mailbox = None
sampler = NeighborSampler(num_nodes=graph.num_nodes, num_layers=num_layers, fanout=fanout,graph_data=sample_graph, workers=10,policy = policy_train, graph_name = "train",local_part=dist.get_rank(),edge_part=DistIndex(graph.eids_mapper).part,node_part=DistIndex(graph.nids_mapper).part,probability=args.probability)
eval_sampler = NeighborSampler(num_nodes=graph.num_nodes, num_layers=num_layers, fanout=fanout,graph_data=eval_sample_graph, workers=10,policy = 'recent', graph_name = "eval",local_part=dist.get_rank(),edge_part=DistIndex(graph.eids_mapper).part,node_part=DistIndex(graph.nids_mapper).part,probability=args.probability)
train_data = torch.masked_select(graph.edge_index,train_mask.to(graph.edge_index.device)).reshape(2,-1)
train_ts = torch.masked_select(graph.ts,train_mask.to(graph.edge_index.device))
print('part {}\n'.format(DistIndex(graph.nids_mapper[train_data]).part))
test_range = torch.arange(0,full_sampler_graph['eids'].shape[0],2)
eval_train_data = torch.masked_select(full_sampler_graph['edge_index'][:,test_range],full_train_mask.to(graph.edge_index.device)).reshape(2,-1)
eval_train_ts = torch.masked_select(full_sampler_graph['ts'][test_range],full_train_mask.to(graph.edge_index.device))
test_data = torch.masked_select(full_sampler_graph['edge_index'][:,test_range],test_mask.to(graph.edge_index.device)).reshape(2,-1)
test_ts = torch.masked_select(full_sampler_graph['ts'][test_range],test_mask.to(graph.edge_index.device))
val_data = torch.masked_select(full_sampler_graph['edge_index'][:,test_range],val_mask.to(graph.edge_index.device)).reshape(2,-1)
val_ts = torch.masked_select(full_sampler_graph['ts'][test_range],val_mask.to(graph.edge_index.device))
train_data = DataSet(edges = train_data,ts =train_ts,eids = torch.nonzero(train_mask).reshape(-1))
eval_train_data = DataSet(edges = eval_train_data,ts = eval_train_ts,eids = full_train_mask.nonzero().reshape(-1))
test_data = DataSet(edges = test_data,ts =test_ts,eids = test_mask.nonzero().reshape(-1))
val_data = DataSet(edges = val_data,ts = val_ts,eids = val_mask.nonzero().reshape(-1))
print('ts {} {} {} {}'.format(train_data.ts,eval_train_data.ts,test_data.ts,val_data.ts))
neg_samples = args.eval_neg_samples
mask = DistIndex(graph.nids_mapper[graph.edge_index[1,:]].to('cpu')).part == dist.get_rank()
if args.local_neg_sample:
print('dst len {} origin len {}'.format(graph.edge_index[1,mask].unique().shape[0],full_dst.unique().shape[0]))
train_neg_sampler = LocalNegativeSampling('triplet',amount = args.neg_samples,dst_node_list = graph.edge_index[1,mask].unique())
else:
train_neg_sampler = LocalNegativeSampling('triplet',amount = args.neg_samples,dst_node_list = full_dst.unique())
print(train_neg_sampler.dst_node_list)
neg_sampler = LocalNegativeSampling('triplet',amount= neg_samples,dst_node_list = full_dst.unique(),seed=12357)
trainloader = DistributedDataLoader(graph,eval_train_data,sampler = sampler,
sampler_fn = SAMPLE_TYPE.SAMPLE_FROM_TEMPORAL_EDGES,
neg_sampler=train_neg_sampler,
batch_size = int(train_param['batch_size'])*dist.get_world_size(),
shuffle=False,
drop_last=True,
chunk_size = None,
mode='train',
queue_size = 200,
mailbox = mailbox,
is_pipeline=False,
use_local_feature = False,
device = torch.device('cuda:{}'.format(local_rank)),
probability=train_cross_probability
)
eval_trainloader = DistributedDataLoader(graph,eval_train_data,sampler = eval_sampler,
sampler_fn = SAMPLE_TYPE.SAMPLE_FROM_TEMPORAL_EDGES,
neg_sampler=neg_sampler,
batch_size = train_param['batch_size'],
shuffle=False,
drop_last=False,
chunk_size = None,
mode='eval_train',
queue_size = 100,
mailbox = mailbox,
device = torch.device('cuda:{}'.format(local_rank))
)
testloader = DistributedDataLoader(graph,test_data,sampler = eval_sampler,
sampler_fn = SAMPLE_TYPE.SAMPLE_FROM_TEMPORAL_EDGES,
neg_sampler=neg_sampler,
batch_size = train_param['batch_size'],
shuffle=False,
drop_last=False,
chunk_size = None,
mode='test',
queue_size = 100,
mailbox = mailbox,
device = torch.device('cuda:{}'.format(local_rank))
)
valloader = DistributedDataLoader(graph,val_data,sampler = eval_sampler,
sampler_fn = SAMPLE_TYPE.SAMPLE_FROM_TEMPORAL_EDGES,
neg_sampler=neg_sampler,
batch_size = train_param['batch_size'],
shuffle=False,
drop_last=False,
chunk_size = None,
train=False,
mode='val',
queue_size = 100,
mailbox = mailbox,
device = torch.device('cuda:{}'.format(local_rank))
)
print('init dataloader')
gnn_dim_node = 0 if graph.nfeat is None else graph.nfeat.shape[1]
gnn_dim_edge = 0 if graph.efeat is None else graph.efeat.shape[1]
avg_time = 0
if use_cuda:
model = GeneralModel(gnn_dim_node, gnn_dim_edge, sample_param, memory_param, gnn_param, train_param,graph.ids.shape[0],mailbox).cuda()
device = torch.device('cuda')
else:
model = GeneralModel(gnn_dim_node, gnn_dim_edge, sample_param, memory_param, gnn_param, train_param,graph.ids.shape[0],mailbox)
device = torch.device('cpu')
model = DDP(model,find_unused_parameters=True)
def count_parameters(model):
return sum(p.numel()*p.element_size()/1024/1024 for p in model.parameters() if p.requires_grad)
print(f'The model has {count_parameters(model):,} trainable parameters')
train_stream = torch.cuda.Stream()
def eval(mode='val'):
model.eval()
aps = list()
aucs_mrrs = list()
if mode == 'val':
loader = valloader
elif mode == 'test':
loader = testloader
elif mode == 'train':
loader = eval_trainloader
err_cnt = 0
err_cross_part = 0
true_cnt = 0
true_cross_cnt = 0
with torch.no_grad():
total_loss = 0
signal = torch.tensor([0],dtype = int,device = device)
for roots,mfgs,metadata in loader:
if ctx.memory_group == 0:
pred_pos, pred_neg = model(mfgs,metadata,neg_samples=neg_samples)
#print('check {}\n'.format(model.module.memory_updater.last_updated_nid))
y_pred = torch.cat([pred_pos, pred_neg], dim=0).sigmoid().cpu()
y_true = torch.cat([torch.ones(pred_pos.size(0)), torch.zeros(pred_neg.size(0))], dim=0)
aps.append(average_precision_score(y_true, y_pred.detach().numpy()))
aucs_mrrs.append(roc_auc_score(y_true, y_pred))
if mailbox is not None:
src = metadata['src_pos_index']
dst = metadata['dst_pos_index']
ts = roots.ts
if graph.efeat is None:
edge_feats = None
elif(graph.efeat.device.type != 'cpu'):
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')]).to('cuda')
#edge_feats = graph.get_dist_efeat(graph.eids_mapper[roots.eids.to('cpu')].to('cuda'),is_sorted = False)#graph.efeat[roots.eids.to('cpu')].to('cuda')
else:
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')])
#edge_feats = graph.get_dist_efeat(graph.eids_mapper[roots.eids.to('cpu')],is_sorted=False)#graph.efeat[roots.eids]
#print(mfgs[0][0].srcdata['ID'])
dist_index_mapper = mfgs[0][0].srcdata['ID']
root_index = torch.cat((src,dst))
#print('{} {} {}'.format((~(dist_index_mapper==model.module.memory_updater.last_updated_nid)).nonzero(),model.module.memory_updater.last_updated_nid,dist_index_mapper))
last_updated_nid = model.module.memory_updater.last_updated_nid[root_index]
last_updated_memory = model.module.memory_updater.last_updated_memory[root_index]
last_updated_ts=model.module.memory_updater.last_updated_ts[root_index]
#print('root shape {} unique {} {}\n'.format(root_index.shape,dist_index_mapper[root_index].unique().shape,last_updated_nid.unique().shape))
index, memory, memory_ts = mailbox.get_update_memory(last_updated_nid,
last_updated_memory,
last_updated_ts,
model.module.embedding)
#print('index {} {}\n'.format(index.shape,dist_index_mapper[torch.cat((src,dst))].unique().shape))
index, mail, mail_ts = mailbox.get_update_mail(dist_index_mapper,
src,dst,ts,edge_feats,
model.module.memory_updater.last_updated_memory,
model.module.embedding,use_src_emb,use_dst_emb,
)
if memory_param['historical_fix'] == True:
mailbox.set_memory_all_reduce(index,memory,memory_ts,mail,mail_ts,reduce_Op = 'max', async_op = False,filter=model.module.memory_updater.filter,set_remote=True,mode='historical')
else:
mailbox.set_memory_all_reduce(index,memory,memory_ts,mail,mail_ts,reduce_Op = 'max', async_op = False,filter=None,set_remote=True,mode='all_reduce')
ap = torch.empty([1])
auc_mrr = torch.empty([1])
if(ctx.memory_group==0):
world_size = dist.get_world_size()
ap[0] = torch.tensor(aps).mean()
auc_mrr[0] = torch.tensor(aucs_mrrs).mean()#float(aucs_mrrs.clone().mean())
print('mode: {} {} {}'.format(mode,ap,auc_mrr))
dist.all_reduce(ap,group = ctx.gloo_group)
ap/=ctx.memory_group_size
dist.all_reduce(auc_mrr,group=ctx.gloo_group)
auc_mrr/=ctx.memory_group_size
dist.broadcast(ap,0,group=ctx.gloo_group)
dist.broadcast(auc_mrr,0,group=ctx.gloo_group)
return ap.item(), auc_mrr.item()
def normalize(x):
if not (x.max().item() == 0):
x = x - x.min()
x = x / x.max()
x = 2*x - 1
return x
def inner_prod(x1,x2):
cos = torch.nn.CosineSimilarity(dim=0)
return cos(normalize(x1),normalize(x2)).sum()/x1.size(dim=0)
creterion = torch.nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=train_param['lr'])
early_stopper = EarlyStopMonitor(max_round=args.patience)
MODEL_SAVE_PATH = f'../saved_models/{args.model}-{args.dataname}-{dist.get_world_size()}.pth'
epoch_cnt = 0
test_ap_list = []
val_list = []
loss_list = []
def fetch_async():
trainloader.async_feature()
for e in range(train_param['epoch']):
model.module.memory_updater.empty_cache()
tt._zero()
torch.cuda.synchronize()
epoch_start_time = time.time()
epoch_cnt = epoch_cnt + 1
train_aps = list()
print('Epoch {:d}:'.format(e))
time_prep = 0
total_loss = 0
model.train()
if mailbox is not None:
mailbox.reset()
model.module.memory_updater.last_updated_nid = None
model.module.memory_updater.last_updated_memory = None
model.module.memory_updater.last_updated_ts = None
sum_local_comm = 0
sum_remote_comm = 0
sum_local_edge_comm = 0
sum_remote_edge_comm = 0
local_access = []
remote_access = []
local_comm = []
remote_comm = []
local_edge_access = []
remote_edge_access = []
local_edge_comm = []
remote_edge_comm = []
b_cnt = 0
for roots,mfgs,metadata in trainloader:
#print('rank is {} batch max ts is {} batch min ts is {}'.format(dist.get_rank(),roots.ts.min(),roots.ts.max()))
b_cnt = b_cnt + 1
local_access.append(trainloader.local_node)
remote_access.append(trainloader.remote_node)
local_edge_access.append(trainloader.local_edge)
remote_edge_access.append(trainloader.remote_edge)
local_comm.append((DistIndex(mfgs[0][0].srcdata['ID']).part == dist.get_rank()).sum().item())
remote_comm.append((DistIndex(mfgs[0][0].srcdata['ID']).part != dist.get_rank()).sum().item())
local_edge_comm.append((DistIndex(mfgs[0][0].edata['ID']).part == dist.get_rank()).sum().item())
remote_edge_comm.append((DistIndex(mfgs[0][0].edata['ID']).part != dist.get_rank()).sum().item())
sum_local_comm +=local_comm[b_cnt-1]
sum_remote_comm +=remote_comm[b_cnt-1]
sum_local_edge_comm +=local_edge_comm[b_cnt-1]
sum_remote_edge_comm +=remote_edge_comm[b_cnt-1]
"""
if mailbox is not None:
if(graph.efeat.device.type != 'cpu'):
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')]).to('cuda')
#edge_feats = graph.get_dist_efeat(graph.eids_mapper[roots.eids.to('cpu')].to('cuda'),is_sorted = False)#graph.efeat[roots.eids.to('cpu')].to('cuda')
else:
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')])
src = metadata['src_pos_index']
dst = metadata['dst_pos_index']
ts = roots.ts
update_mail = True
param = (update_mail,src,dst,ts,edge_feats,fetch_async)
else:
param = None
"""
param = None
model.train()
optimizer.zero_grad()
pred_pos, pred_neg = model(mfgs,metadata,neg_samples=args.neg_samples,async_param = param)
loss_start =tt.start_gpu()
if memory_param['historical_fix'] == True:
loss = creterion(pred_pos, torch.ones_like(pred_pos)) + 0.1*inner_prod(model.module.memory_updater.update_memory,model.module.memory_updater.prev_memory)
else:
loss = creterion(pred_pos, torch.ones_like(pred_pos))
loss += creterion(pred_neg, torch.zeros_like(pred_neg))
total_loss += float(loss)
#mailbox.handle_last_async()
#trainloader.async_feature()
#torch.cuda.synchronize()
loss.backward()
optimizer.step()
back_end = tt.elapsed_event(loss_start)
tt.time_backward+=back_end
#torch.cuda.synchronize()
## train aps
#y_pred = torch.cat([pred_pos, pred_neg], dim=0).sigmoid().cpu()
#y_true = torch.cat([torch.ones(pred_pos.size(0)), torch.zeros(pred_neg.size(0))], dim=0)
#train_aps.append(average_precision_score(y_true, y_pred.detach().numpy()))
#torch.cuda.synchronize()
#mailbox.update_shared()
#mailbox.update_p2p()
if mailbox is not None:
t_other = tt.start_gpu()
if(graph.efeat.device.type != 'cpu'):
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')]).to('cuda')
#edge_feats = graph.get_dist_efeat(graph.eids_mapper[roots.eids.to('cpu')].to('cuda'),is_sorted = False)#graph.efeat[roots.eids.to('cpu')].to('cuda')
else:
edge_feats = graph.get_local_efeat(graph.eids_mapper[roots.eids.to('cpu')])
src = metadata['src_pos_index']
dst = metadata['dst_pos_index']
ts = roots.ts
dist_index_mapper = mfgs[0][0].srcdata['ID']
root_index = torch.cat((src,dst))
#print('{} {} {}'.format((~(dist_index_mapper==model.module.memory_updater.last_updated_nid)).nonzero(),model.module.memory_updater.last_updated_nid,dist_index_mapper))
last_updated_nid = model.module.memory_updater.last_updated_nid[root_index]
last_updated_memory = model.module.memory_updater.last_updated_memory[root_index]
last_updated_ts=model.module.memory_updater.last_updated_ts[root_index]
#print('root shape {} unique {} {}\n'.format(root_index.shape,dist_index_mapper[root_index].unique().shape,last_updated_nid.unique().shape))
index, memory, memory_ts = mailbox.get_update_memory(last_updated_nid,
last_updated_memory,
last_updated_ts,
model.module.embedding)
#print('index {} {}\n'.format(index.shape,dist_index_mapper[torch.cat((src,dst))].unique().shape))
index, mail, mail_ts = mailbox.get_update_mail(dist_index_mapper,
src,dst,ts,edge_feats,
model.module.memory_updater.last_updated_memory,
model.module.embedding,use_src_emb,use_dst_emb,
)
t_end = tt.elapsed_event(t_other)
t_update_memory = tt.start_gpu()
if memory_param['historical'] == True:
mailbox.set_memory_all_reduce(index,memory,memory_ts,mail,mail_ts,reduce_Op = 'max', async_op = False,filter=model.module.memory_updater.filter,set_remote=True,mode='historical')
else:
mailbox.set_memory_all_reduce(index,memory,memory_ts,mail,mail_ts,reduce_Op = 'max', async_op = False,filter=None,set_remote=True,mode='all_reduce')
t_end_update = tt.elapsed_event(t_update_memory)
tt.time_memory_sync += t_end_update
torch.cuda.synchronize()
time_prep = time.time() - epoch_start_time
avg_time += time.time() - epoch_start_time
train_ap = float(torch.tensor(train_aps).mean())
print('\ttrain time:{:.2f}s\n'.format(time_prep))
print(trainloader.local_node)
local_node=torch.tensor([trainloader.local_node])
remote_node=torch.tensor([trainloader.remote_node])
local_edge=torch.tensor([trainloader.local_edge])
remote_edge=torch.tensor([trainloader.remote_edge])
tot_comm_count=torch.tensor([mailbox.tot_comm_count])
tot_shared_count=torch.tensor([mailbox.tot_shared_count])
torch.distributed.all_reduce(local_node,group=ctx.gloo_group)
torch.distributed.all_reduce(remote_node,group=ctx.gloo_group)
torch.distributed.all_reduce(local_edge,group=ctx.gloo_group)
torch.distributed.all_reduce(remote_edge,group=ctx.gloo_group)
torch.distributed.all_reduce(tot_comm_count,group=ctx.gloo_group)
torch.distributed.all_reduce(tot_shared_count,group=ctx.gloo_group)
print('local node number {} remote node number {} local edge {} remote edge{}\n'.format(local_node,remote_node,local_edge,remote_edge))
print(' comm local node number {} remote node number {} local edge {} remote edge{}\n'.format(sum_local_comm,sum_remote_comm,sum_local_edge_comm,sum_remote_edge_comm))
print('memory comm {} shared comm {}\n'.format(tot_comm_count,tot_shared_count))
if(e==0):
torch.save((local_access,remote_access,local_edge_access,remote_edge_access,local_comm,remote_comm,local_edge_comm,remote_edge_comm),'all/{}/comm/comm_{}_{}_{}_{}_{}_{}_{}_{}.pt'.format(args.dataname,args.partition,args.topk,dist.get_world_size(),dist.get_rank(),args.sample_type,args.probability,args.memory_type,args.shared_memory_ssim))
tt.print()
tt._zero()
ap = 0
auc = 0
ap, auc = eval('val')
test_ap,test_auc = eval('test')
test_ap_list.append((test_ap,test_auc))
early_stop = early_stopper.early_stop_check(ap)
trainloader.local_node = 0
trainloader.remote_node = 0
trainloader.local_edge = 0
trainloader.remote_edge = 0
mailbox.tot_comm_count = 0
mailbox.tot_shared_count = 0
value,counts = torch.unique(graph.edge_index.reshape(-1),return_counts = True)
node_degree = torch.zeros(graph.num_nodes,dtype=torch.long)
value = value.to('cpu')
counts = counts.to('cpu')
node_degree[value] = counts
if dist.get_world_size()==1:
mailbox.mon.draw(node_degree,args.dataname,e)
mailbox.mon.set_zero()
#mailbox.mon.draw(node_degree,args.dataname,e)
#mailbox.mon.set_zero()
loss_list.append(total_loss)
val_list.append(ap)
if early_stop:
dist.barrier()
print("Early stopping at epoch {:d}\n".format(e))
print(f"Loading the best model at epoch {early_stopper.best_epoch}\n")
best_model_path = get_checkpoint_path(early_stopper.best_epoch)
model.module.load_state_dict(torch.load(best_model_path))
break
else:
print('\ttrain loss:{:.4f} train ap:{:4f} val ap:{:4f} val auc:{:4f} test ap {:4f} test auc{:4f}\n'.format(total_loss,train_ap, ap, auc,test_ap,test_auc))
print('\ttotal time:{:.2f}s prep time:{:.2f}s\n'.format(time.time()-epoch_start_time, time_prep))
torch.save(model.module.state_dict(), get_checkpoint_path(e))
torch.save(val_list,'all/{}/val_{}_{}_{}_{}_{}_{}_{}_{}.pt'.format(args.dataname,args.partition,args.topk,dist.get_world_size(),dist.get_rank(),args.sample_type,args.probability,args.memory_type,args.shared_memory_ssim))
torch.save(loss_list,'all/{}/loss_{}_{}_{}_{}_{}_{}_{}_{}.pt'.format(args.dataname,args.partition,args.topk,dist.get_world_size(),dist.get_rank(),args.sample_type,args.probability,args.memory_type,args.shared_memory_ssim))
print(avg_time)
if not early_stop:
dist.barrier()
print(f"Loading the best model at epoch {early_stopper.best_epoch}")
best_model_path = get_checkpoint_path(early_stopper.best_epoch)
model.module.load_state_dict(torch.load(best_model_path))
print('best test AP:{:4f} test auc{:4f}'.format(*test_ap_list[early_stopper.best_epoch]))
val_list = torch.tensor(val_list)
loss_list = torch.tensor(loss_list)
print('test_dataset {} avg_time {} \n'.format(test_data.edges.shape[1],avg_time/epoch_cnt))
torch.save(model.module.state_dict(), MODEL_SAVE_PATH)
ctx.shutdown()
if __name__ == "__main__":
main()
starrygl/module/modules.py
...@@ -114,14 +114,13 @@ class GeneralModel(torch.nn.Module): ...@@ -114,14 +114,13 @@ class GeneralModel(torch.nn.Module):
def forward(self, mfgs, metadata = None,neg_samples=1, mode = 'triplet',async_param = None): def forward(self, mfgs, metadata = None,neg_samples=1, mode = 'triplet',async_param = None):
t0 = time.time() t0 = tt.start_gpu()
if self.memory_param['type'] == 'node': if self.memory_param['type'] == 'node':
self.memory_updater(mfgs[0],async_param) self.memory_updater(mfgs[0],async_param)
t_mem = tt.elapsed_event(t0)
tt.time_memory_updater += t_mem
out = list() out = list()
start = torch.cuda.Event(enable_timing=True) t1 = tt.start_gpu()
end = torch.cuda.Event(enable_timing=True)
start.record()
for l in range(self.gnn_param['layer']): for l in range(self.gnn_param['layer']):
for h in range(self.sample_param['history']): for h in range(self.sample_param['history']):
rst = self.layers['l' + str(l) + 'h' + str(h)](mfgs[l][h]) rst = self.layers['l' + str(l) + 'h' + str(h)](mfgs[l][h])
...@@ -145,7 +144,10 @@ class GeneralModel(torch.nn.Module): ...@@ -145,7 +144,10 @@ class GeneralModel(torch.nn.Module):
#print('time {}\n'.format(elapsed_time_ms)) #print('time {}\n'.format(elapsed_time_ms))
#print('pos src {} \n pos dst {} \n neg dst{} \n'.format(h_pos_src, h_pos_dst,h_neg_dst)) #print('pos src {} \n pos dst {} \n neg dst{} \n'.format(h_pos_src, h_pos_dst,h_neg_dst))
#print('pre predict {}'.format(mfgs[0][0].srcdata['ID'])) #print('pre predict {}'.format(mfgs[0][0].srcdata['ID']))
return self.edge_predictor(h_pos_src, h_pos_dst, None , h_neg_dst, neg_samples=neg_samples, mode = mode) pred = self.edge_predictor(h_pos_src, h_pos_dst, None , h_neg_dst, neg_samples=neg_samples, mode = mode)
t_embedding = tt.elapsed_event(t1)
tt.time_embedding+=t_embedding
return pred
class NodeClassificationModel(torch.nn.Module): class NodeClassificationModel(torch.nn.Module):
......
starrygl/sample/data_loader.py
...@@ -23,7 +23,7 @@ import torch.distributed as dist ...@@ -23,7 +23,7 @@ import torch.distributed as dist
from torch_geometric.data import Data from torch_geometric.data import Data
import os.path as osp import os.path as osp
import math import math
from starrygl.sample.count_static import time_count as tt
import data_loader as data_loader import data_loader as data_loader
import concurrent.futures import concurrent.futures
executor = concurrent.futures.ThreadPoolExecutor(max_workers=1) executor = concurrent.futures.ThreadPoolExecutor(max_workers=1)
...@@ -294,6 +294,7 @@ class DistributedDataLoader: ...@@ -294,6 +294,7 @@ class DistributedDataLoader:
#with torch.cuda.stream(stream): #with torch.cuda.stream(stream):
if self.is_pipeline is False: if self.is_pipeline is False:
if self.recv_idxs < self.expected_idx: if self.recv_idxs < self.expected_idx:
t0 = tt.start_gpu()
data = self._next_data() data = self._next_data()
batch_data,dist_nid,dist_eid = graph_sample( batch_data,dist_nid,dist_eid = graph_sample(
self.graph, self.graph,
...@@ -305,9 +306,14 @@ class DistributedDataLoader: ...@@ -305,9 +306,14 @@ class DistributedDataLoader:
eid_mapper = self.graph.eids_mapper) eid_mapper = self.graph.eids_mapper)
root,mfgs,metadata = batch_data root,mfgs,metadata = batch_data
t_sample = tt.elapsed_event(t0)
tt.time_sample_and_build+=t_sample
t1 = tt.start_gpu()
edge_feat = get_edge_feature_by_dist(self.graph,dist_eid,is_local,out_device=self.device) edge_feat = get_edge_feature_by_dist(self.graph,dist_eid,is_local,out_device=self.device)
node_feat,mem = get_node_feature_by_dist(self.graph,self.mailbox,dist_nid, is_local,out_device=self.device) node_feat,mem = get_node_feature_by_dist(self.graph,self.mailbox,dist_nid, is_local,out_device=self.device)
prepare_input(node_feat,edge_feat,mem,mfgs,dist_nid,dist_eid) prepare_input(node_feat,edge_feat,mem,mfgs,dist_nid,dist_eid)
t_fetch = tt.elapsed_event(t1)
tt.time_memory_fetch += t_fetch
#if(self.mailbox is not None and self.mailbox.historical_cache is not None): #if(self.mailbox is not None and self.mailbox.historical_cache is not None):
# id = batch_data[1][0][0].srcdata['ID'] # id = batch_data[1][0][0].srcdata['ID']
# mask = DistIndex(id).is_shared # mask = DistIndex(id).is_shared
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment