Adding the RecurrentGCN models

RecurrentGCN

Author

SEOYEON CHOI

Published

May 27, 2023

RecurrentGCN

Import

import itstgcntry
import torch
import itstgcntry.planner

Try

import pandas as pd

pd.read_csv('./simulation_results/2023-05-29_11-18-32.csv').groupby(['RecurrentGCN','method','mrate'])['mse'].mean().reset_index()

	RecurrentGCN	method	mrate	mse
0	DCRNN	IT-STGCN	0.8	1.517346
1	DCRNN	STGCN	0.8	1.521906
2	EvolveGCNH	IT-STGCN	0.8	1.338225
3	EvolveGCNH	STGCN	0.8	1.337022
4	EvolveGCNO	IT-STGCN	0.8	1.370455
5	EvolveGCNO	STGCN	0.8	1.326990
6	GCLSTM	IT-STGCN	0.8	1.529536
7	GCLSTM	STGCN	0.8	1.600452
8	GConvGRU	IT-STGCN	0.8	1.567661
9	GConvGRU	STGCN	0.8	1.654162
10	GConvLSTM	IT-STGCN	0.8	1.494284
11	GConvLSTM	STGCN	0.8	1.633645
12	LRGCN	IT-STGCN	0.8	1.512042
13	LRGCN	STGCN	0.8	1.604186
14	MPNNLSTM	IT-STGCN	0.8	1.386678
15	MPNNLSTM	STGCN	0.8	1.351526
16	TGCN	IT-STGCN	0.8	1.279269
17	TGCN	STGCN	0.8	1.278411

pd.read_csv('./simulation_results/2023-05-29_12-31-06.csv').groupby(['RecurrentGCN','method','mrate'])['mse'].mean().reset_index()

	RecurrentGCN	method	mrate	mse
0	DCRNN	IT-STGCN	0.6	1.457874
1	DCRNN	STGCN	0.6	1.523312
2	EvolveGCNH	IT-STGCN	0.6	1.265695
3	EvolveGCNH	STGCN	0.6	1.290008
4	EvolveGCNO	IT-STGCN	0.6	1.288295
5	EvolveGCNO	STGCN	0.6	1.308886
6	GCLSTM	IT-STGCN	0.6	1.304539
7	GCLSTM	STGCN	0.6	1.537681
8	GConvGRU	IT-STGCN	0.6	1.402607
9	GConvGRU	STGCN	0.6	1.582745
10	GConvLSTM	IT-STGCN	0.6	1.299263
11	GConvLSTM	STGCN	0.6	1.534498
12	LRGCN	IT-STGCN	0.6	1.295239
13	LRGCN	STGCN	0.6	1.527917
14	MPNNLSTM	IT-STGCN	0.6	1.287649
15	MPNNLSTM	STGCN	0.6	1.307589
16	TGCN	IT-STGCN	0.6	1.276318
17	TGCN	STGCN	0.6	1.262256

GConvGRU

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

/home/csy/Dropbox/blog/posts/GCN/itstgcntry/utils.py:72: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /opt/conda/conda-bld/pytorch_1682343998658/work/torch/csrc/utils/tensor_new.cpp:245.)
  lags = torch.tensor(train_dataset.features).shape[-1]

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.GConvGRU_RecurrentGCN(train_dataset_padded,filters=1)

lrnr.learn(model,epoch=5)

5/5

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.GConvGRU_RecurrentGCN(train_dataset_padded,filters=1)

lrnr1.learn(model1,epoch=5)

5/5

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

DCRNN

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.DCRNN_RecurrentGCN(train_dataset_padded,filters=1)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.DCRNN_RecurrentGCN(train_dataset_padded,filters=1)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

EvolveGCNH

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.EvolveGCNH_RecurrentGCN(train_dataset_padded)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.EvolveGCNH_RecurrentGCN(train_dataset_padded)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

EvolveGCNO

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.EvolveGCNO_RecurrentGCN(train_dataset_padded)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.EvolveGCNO_RecurrentGCN(train_dataset_padded)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

GCLSTM

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.GCLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.GCLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

GConvLSTM

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.GConvLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.GConvLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

LRGCN

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.LRGCN_RecurrentGCN(train_dataset_padded, filters=1)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.LRGCN_RecurrentGCN(train_dataset_padded, filters=1)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

MPNNLSTM

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.MPNNLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.MPNNLSTM_RecurrentGCN(train_dataset_padded, filters=1)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

TGCN

data_dict = itstgcntry.load_data('./data/fivenodes.pkl')
loader = itstgcntry.DatasetLoader(data_dict)

dataset = loader.get_dataset(lags=2)
train_dataset, test_dataset = itstgcntry.temporal_signal_split(dataset, train_ratio=0.8)

mindex = [list(range(10,100)),[],list(range(50,80)),[],[]]
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='block')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

mindex = itstgcntry.rand_mindex(train_dataset,mrate=0.5)
train_dataset_miss = itstgcntry.miss(train_dataset,mindex,mtype='rand')
train_dataset_padded = itstgcntry.padding(train_dataset_miss) # padding(train_dataset_miss,method='linear'와 같음)

- 학습

lrnr = itstgcntry.StgcnLearner(train_dataset_padded)

model = itstgcntry.TGCN_RecurrentGCN(train_dataset_padded, filters=32)

lrnr.learn(model,epoch=50)

50/50

lrnr1 = itstgcntry.ITStgcnLearner(train_dataset_padded)

model1 = itstgcntry.TGCN_RecurrentGCN(train_dataset_padded, filters=32)

lrnr1.learn(model1,epoch=50)

50/50

- 적합값

# lrnr(train_dataset_padded) 
# lrnr(test_dataset)['yhat'].shape

실행하면 X,y,yhat 출력

- 모형 평가 및 시각화

evtor = itstgcntry.Evaluator(lrnr,train_dataset_padded,test_dataset)

fig = evtor.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig

evtor1 = itstgcntry.Evaluator(lrnr1,train_dataset_padded,test_dataset)

fig = evtor1.plot('--.',h=5,max_node=3,label='complete data',alpha=0.5) # max_nodes 는 1보다 커야함
fig.set_figwidth(12)
fig.set_figheight(12)
fig.tight_layout()
fig