Skip to content

Commit 9785eab

Browse files
thieu1995thieu1995
committed
Update project
1 parent 187aaa8 commit 9785eab

File tree

2 files changed

+239
-60
lines changed

2 files changed

+239
-60
lines changed

‎project_note.md‎

Lines changed: 83 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,83 @@
1+
# Links:
2+
```code
3+
1. code: https://github.com/chasebk/code_OTWO_ELM
4+
2. paper: https://github.com/chasebk/paper_OTWO_ELM
5+
```
6+
7+
# Model comparison (normal and stability)
8+
1. MLNN
9+
2. ELM
10+
3. GA-ELM
11+
4. PSO-ELM
12+
5. TWO-ELM
13+
6. OTWO-ELM
14+
15+
16+
# Project structure
17+
1. General view class
18+
![Our model](paper/images/code/all_code_wrapper.png)
19+
20+
2. Details view class
21+
* root files
22+
23+
![](paper/images/code/root_ann.png) ![](paper/images/code/root_rnn.png) ![](paper/images/code/root_hybrid_mlnn.png)
24+
25+
* algorithm files
26+
27+
![](paper/images/code/GA.png) ![](paper/images/code/DE.png) ![](paper/images/code/PSO.png)
28+
29+
![](paper/images/code/CRO.png) ![](paper/images/code/BFO.png)
30+
31+
* main files
32+
33+
![Our model](paper/images/code/hybrid_mlnn.png)
34+
35+
![Our model](paper/images/code/neural_network.png)
36+
37+
38+
39+
40+
## Optimized the number of hidden neuron in hidden layer
41+
```code
42+
https://iopscience.iop.org/article/10.1088/1757-899X/261/1/012002/pdf
43+
https://sci-hub.tw/10.1016/j.neucom.2011.12.062
44+
https://arxiv.org/pdf/1501.06115.pdf
45+
https://www.researchgate.net/publication/271201213_A_detailed_analysis_on_extreme_learning_machine_and_novel_approaches_based_on_ELM
46+
https://www.researchgate.net/post/Can_anyone_help_with_Training_Extreme_Learning_Machines_neural_network_on_very_Large_Datasets
47+
https://www.sciencedirect.com/science/article/pii/S2212671612002181
48+
http://www3.ntu.edu.sg/home/ASAHTan/Papers/2008/ELM%20NEUCOM%2008.pdf
49+
http://or.nsfc.gov.cn/bitstream/00001903-5/112878/1/1000004116461.pdf
50+
51+
```
52+
53+
54+
## Server Errors
55+
```code
56+
1. Maybe multi-threading error between numpy and openBlas
57+
- We should check what is the core-backend of numpy in server: blas or mkl
58+
python
59+
import numpy
60+
numpy.__config__.show()
61+
62+
https://stackoverflow.com/questions/17053671/python-how-do-you-stop-numpy-from-multithreading
63+
https://stackoverflow.com/questions/19257070/unintented-multithreading-in-python-scikit-learn
64+
65+
- Numpy will consume so much time when running multi-thread because the time for exchanging data among cores.
66+
So we should prevent numpy running multi-thread on server.
67+
68+
Adding to the file: ~/.bashrc or ~/.bash_profile:
69+
export OPENBLAS_NUM_THREADS=1 (If OpenBlas)
70+
export MKL_NUM_THREADS=1 (If MKL)
71+
72+
export OPENBLAS_NUM_THREADS=1
73+
export MKL_NUM_THREADS=1
74+
75+
source ~/.bashrc
76+
77+
2. Cannot share object related to matplotlib on server
78+
sudo apt update
79+
sudo apt install libgl1-mesa-glx
80+
```
81+
82+
83+

‎readme.md‎

Lines changed: 156 additions & 60 deletions
Original file line numberDiff line numberDiff line change
@@ -1,87 +1,183 @@
1-
# Links:
2-
```code
3-
1. git: https://github.com/thieunguyen5991/two_elm/
4-
2. paper: https://bitbucket.org/nguyenthieu2102/paper_two/
5-
```
1+
# A Novel Workload Prediction Model Using ELM with Opposition-based Tug of War Optimization
2+
3+
## Cite us
4+
* If you see my code and data useful and use it, please cite us as follows
5+
6+
* Nguyen, T., Nguyen, T., Nguyen, B. M., & Nguyen, G. (2019). Efficient Time-Series Forecasting Using Neural Network and Opposition-Based Coral Reefs Optimization. International Journal of Computational Intelligence Systems, 12(2), 1144-1161.
7+
8+
* Nguyen, T., Nguyen, B. M., & Nguyen, G. (2019, April). Building Resource Auto-scaler with Functional-Link Neural Network and Adaptive Bacterial Foraging Optimization. In International Conference on Theory and Applications of Models of Computation (pp. 501-517). Springer, Cham.
9+
10+
* Nguyen, T., Tran, N., Nguyen, B. M., & Nguyen, G. (2018, November). A Resource Usage Prediction System Using Functional-Link and Genetic Algorithm Neural Network for Multivariate Cloud Metrics. In 2018 IEEE 11th Conference on Service-Oriented Computing and Applications (SOCA) (pp. 49-56). IEEE.
11+
12+
* If you want to know more about code, or want a pdf of both above paper, contact me: [email protected]
13+
14+
* Take a look at this repos, the simplify code using python (numpy) for all algorithms above. (without neural networks)
15+
16+
* https://github.com/thieunguyen5991/metaheuristics
17+
18+
## How to read my repository
19+
1. data: include raw and formatted data
20+
2. envs: include conda environment and how to install conda environment
21+
3. utils: Helped functions such as IO, Draw, Math, Settings (for all model and parameters), Preprocessing...
22+
4. paper: include 2 main folders:
23+
* results: forecasting results of all models (3 folders inside)
24+
* final: final forecasting results (runs on server)
25+
* test: testing forecasting results (runs on personal computer, just for test)
26+
* temp: nothing (just for copy of test folder)
27+
* scaling: scaling results
28+
5. model: (4 folders)
29+
* root: (want to understand the code, read this classes first)
30+
* root_base.py: root for all models (traditional, hybrid and variants...)
31+
* root_algo.py: root for all optimization algorithms
32+
* traditional: root for all traditional models (inherit: root_base)
33+
* hybrid: root for all hybrid models (inherit: root_base)
34+
* optimizer: (this classes inherit: root_algo.py)
35+
* evolutionary: include algorithms related to evolution algorithm such as GA, DE,..
36+
* swarm: include algorithms related to swarm optimization such as PSO, BFO, ...
37+
* physics: include algorithms related to physics optimization such as QSO, TWO, ...
38+
* main: (final models)
39+
* this classes will use those optimizer above and those root (traditional, hybrid) above
40+
* the running files (outside with the downloaded folder) will call this classes
41+
* the traditional models will use single file such as: traditional_elm, traditional_rnn,...
42+
* the hybrid models will use 2 files, example: hybrid_elm.py and GA.py (optimizer files)
43+
6. special files
44+
* *_scipt.py: running files (*: represent model) such as ga_elm_script.py => GA + ELM
45+
46+
47+
## Notes
648

7-
## Model comparison
8-
1. MLNN
9-
2. ELM
10-
3. GA-ELM
11-
4. PSO-ELM
12-
5. TWO-ELM
13-
6. ITWO-ELM
49+
1. To improve the speed of Pycharm when opening (because Pycharm will indexing when opening), you should right click to
50+
paper and data folder => Mark Directory As => Excluded
1451

52+
2. How to run models?
53+
```code
54+
1. Before runs the models, make sure you clone this repository in your laptop:
55+
https://github.com/chasebk/code_OTWO_ELM
1556
57+
2. Then open it in your editor like Pycharm or Spider...
1658
17-
=== Running ====
18-
1. MLNN (normal mode)
59+
3. Now you need to create python environment using conda (assumpted that you have already had it). Open terminal
60+
conda your_environment_name create -f envs/env.yml (go to the root project folder and create new environment from my file in: envs/env.yml)
1961
20-
==> Last: MLNN (stability mode)
62+
4. Now you can activate your environment and run the models
63+
conda activate your_environment_name # First, activate your environment to get the needed libraries.
64+
python model_name
2165
66+
For example:
67+
conda ai_env create -f envs/env.yml
68+
conda activate ai_env
69+
python elm_script.py
2270
71+
5. My model name:
2372
73+
1. MLNN (1 HL) => mlnn1hl_script.py
74+
2. ELM => elm_script.py
75+
3. GA-ELM => ga_elm_script.py
76+
4. PSO-ELM => pso_elm_script.py
77+
5. ABFOLS-ELM => abfols_elm_script.py
78+
6. DE-ELM => de_elm_script.py
79+
7. TWO-ELM => two_elm_script.py
80+
8. OTWO-ELM => otwo_elm_script.py
81+
```
2482

25-
## Optimized the number of hidden neuron in hidden layer
83+
3. In paper/results/final model includes folder's name represent the data such as
2684
```code
27-
https://iopscience.iop.org/article/10.1088/1757-899X/261/1/012002/pdf
28-
https://sci-hub.tw/10.1016/j.neucom.2011.12.062
29-
https://arxiv.org/pdf/1501.06115.pdf
30-
https://www.researchgate.net/publication/271201213_A_detailed_analysis_on_extreme_learning_machine_and_novel_approaches_based_on_ELM
31-
https://www.researchgate.net/post/Can_anyone_help_with_Training_Extreme_Learning_Machines_neural_network_on_very_Large_Datasets
32-
https://www.sciencedirect.com/science/article/pii/S2212671612002181
33-
http://www3.ntu.edu.sg/home/ASAHTan/Papers/2008/ELM%20NEUCOM%2008.pdf
34-
http://or.nsfc.gov.cn/bitstream/00001903-5/112878/1/1000004116461.pdf
35-
85+
cpu: input model would be cpu, output model would be cpu
86+
ram: same as cpu
87+
multi_cpu : input model would be cpu and ram, output model would be cpu
88+
multi_ram : input model would be cpu and ram, output model would be ram
89+
multi : input model would be cpu and ram, output model would be cpu and ram
3690
```
3791

92+
4. How to change model's parameters?
93+
```code
94+
You can change the model's parameters in file: utils/SettingPaper.py
95+
96+
For example:
97+
98+
+) For traditional models: MLNN, ELM
99+
100+
####: MLNN-1HL
101+
mlnn1hl_paras_final = {
102+
"sliding": [2, 5, 10],
103+
"hidden_sizes" : [[5] ],
104+
"activations": [("elu", "elu")], # 0: elu, 1:relu, 2:tanh, 3:sigmoid
105+
"learning_rate": [0.0001],
106+
"epoch": [5000],
107+
"batch_size": [128],
108+
"optimizer": ["adam"], # GradientDescentOptimizer, AdamOptimizer, AdagradOptimizer, AdadeltaOptimizer
109+
"loss": ["mse"]
110+
}
111+
112+
- If you want to tune the parameters, you can adding more value in each parameters like this:
113+
114+
- sliding: [1, 2, 3, 4] or you want just 1 parameter: [12]
115+
- hidden_sizes: [ [5], [10], [100] ] or [ [14] ]
116+
- activations: [ ("elu", "relu"), ("elu", "tanh") ] or just: [ ("elu", "elu") ]
117+
- learning_rate: [0.1, 0.01, 0.001] or just: [0.1]
118+
....
119+
120+
121+
+ For hybrid models: GA_ELM, PSO_ELM, ABFOLS-ELM, TWO-ELM, OTWO-ELM
122+
123+
#### : GA-ELM
124+
ga_elm_paras_final = {
125+
"sliding": [2, 3, 5],
126+
"hidden_size" : [(5, False)],
127+
"activation": [0], # 0: elu, 1:relu, 2:tanh, 3:sigmoid
128+
"train_valid_rate": [(0.6, 0.4)],
129+
130+
"epoch": [100],
131+
"pop_size": [20], # 100 -> 900
132+
"pc": [0.95], # 0.85 -> 0.97
133+
"pm": [0.025], # 0.005 -> 0.10
134+
"domain_range": [(-1, 1)] # lower and upper bound
135+
}
136+
137+
- Same as traditional models.
138+
```
38139

140+
5. Where is the results folder?
141+
```code
142+
- Look at the running file, for example: ga_elm_script.py
39143
40-
# Project structure
41-
1. General view class
42-
![Our model](paper/images/code/all_code_wrapper.png)
43-
44-
2. Details view class
45-
* root files
46-
47-
![](paper/images/code/root_ann.png) ![](paper/images/code/root_rnn.png) ![](paper/images/code/root_hybrid_mlnn.png)
144+
+) For normal runs (Only run 1 time for each model).
145+
_There are 3 type of results file include: model_name.csv file, model_name.png file and Error-model_name.csv file
146+
_model_name.csv included: y_true and y_predict
147+
_model_name.png is visualized of: y_true and y_predict (test dataset)
148+
_Error-model_name.csv included errors of training dataset after epoch. 1st, 2nd column are: MSE, MAE errors
48149
49-
* algorithm files
150+
=> All 3 type of files above is automatically generated in folder: paper/results/final/...
50151
51-
![](paper/images/code/GA.png) ![](paper/images/code/DE.png) ![](paper/images/code/PSO.png)
152+
+) For stability runs (Run each model 15 times with same parameters).
153+
Because in this test, we don't need to visualize the y_true and y_predict and also don't need to save y_true and y_predict
154+
So I just save 15 times running in the same csv files in folder: paper/results/stability/...
52155
53-
![](paper/images/code/CRO.png) ![](paper/images/code/BFO.png)
156+
- Noted:
54157
55-
* main files
158+
+ In the training set we can use MSE or MAE error. But for the testing set, we can use so much more error like: R2, MAPE, ...
159+
You can find the function code described it in file:
160+
model/root/root_base.py
161+
+ _save_results__: this is for normal runs
162+
+ _save_results_ntimes_run__: this is for stability runs.
56163
57-
![Our model](paper/images/code/hybrid_mlnn.png)
164+
+ You want to change the directory to save results?. Simple.
58165
59-
![Our model](paper/images/code/neural_network.png)
166+
For example: Look at running file: ga_elm_script.py
60167
168+
_Line 14: pathsave = "paper/results/final/" --> This is general path to save file.
169+
But for each dataset like: wc, uk, eu,... I need to save it in different folder right. So now look at
61170
171+
_Line 34: "path_save_result": pathsave + requirement_variables[4].
172+
Here, the requirement_variables[4]: is inside file: utils/SettingPaper.py ("eu/", # path_save_result)
62173
174+
So the final path to save the results for normal runs is: paper/results/final/eu/
175+
```
63176

64-
## Server Errors: Check lỗi multi-threading giữa numpy và openBlas
65-
Ta phải check xem core-backend của numpy nó đang dùng thư viện hỗ trợ nào : blas hay mkl
66-
python
67-
import numpy
68-
numpy.__config__.show()
69-
70-
https://stackoverflow.com/questions/17053671/python-how-do-you-stop-numpy-from-multithreading
71-
https://stackoverflow.com/questions/19257070/unintented-multithreading-in-python-scikit-learn
177+
6. Take a look at project_structure.md file. Describe how the project was built.
72178

73-
---> Để chặn numpy không chạy multi-thread sẽ tốn thời gian trao đổi:
74-
Thêm vào file ~/.bashrc hoặc ~/.bash_profile dòng sau:
75-
export OPENBLAS_NUM_THREADS=1 (Nếu dùng OpenBlas)
76-
export MKL_NUM_THREADS=1 (Nếu dùng MKL)
77179

78-
export OPENBLAS_NUM_THREADS=1
79-
export MKL_NUM_THREADS=1
80-
81-
source ~/.bashrc
180+
## License
82181

83-
## Neu bi loi: Cannot share object lien quan den matplotlib tren server thi sua nhu sau:
84-
```
85-
sudo apt update
86-
sudo apt install libgl1-mesa-glx
87-
```
182+
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
183+

0 commit comments

Comments
 (0)