An open API service indexing awesome lists of open source software.

https://github.com/oniani/cancer_research_gnn


https://github.com/oniani/cancer_research_gnn

Last synced: about 1 year ago
JSON representation

Awesome Lists containing this project

README

          

# Results

- [GraphSAGE](#graphsage)
- [Mean Aggregator](#mean-aggregator)
- [GCN Aggregator](#gcn-aggregator)
- [MoNet](#monet)
- [GAT](#gat)
- [GCN](#gcn)
- [APPNP](#appnp)
- [GIN](#gin)
- [TAGCN](#tagcn)
- [SGC](#sgc)
- [AGNN](#agnn)
- [ChebNet](#chebnet)

## GraphSAGE

### Mean Aggregator

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.9012345679012346 |
| Precision | 0.9261904761904762 |
| Recall | 0.9102607709750566 |
| F-Score | 0.9147043432757718 |

| Hyperparameter | Value |
| --------------------------- | ------------- |
| Dropout probability | 0.25 |
| Learning rate | 1e-2 (0.01) |
| Number of training epochs | 800 |
| Number of hidden gcn units | 16 |
| Number of hidden gcn layers | 1 |
| Weight for L2 loss | 5e-4 (0.0005) |
| Aggregator type | mean |

### GCN Aggregator

StatisticsHyperparameters

| Statistic | Value |
| --------- | -------------------- |
| Accuracy | 0.2222222222222222 |
| Precision | 0.031746031746031744 |
| Recall | 0.14285714285714285 |
| F-Score | 0.051948051948051945 |

| Hyperparameter | Value |
| --------------------------- | ------------- |
| Dropout probability | 0.25 |
| Learning rate | 1e-1 (0.1) |
| Number of training epochs | 800 |
| Number of hidden gcn units | 2 |
| Number of hidden gcn layers | 1 |
| Weight for L2 loss | 5e-4 (0.0005) |
| Aggregator type | gcn |

## MoNet

StatisticsHyperparameters

| Statistic | Value |
| --------- | -------------------- |
| Accuracy | 0.2222222222222222 |
| Precision | 0.031746031746031744 |
| Recall | 0.14285714285714285 |
| F-Score | 0.051948051948051945 |

| Hyperparameter | Value |
| -------------------------------------------------------------------- | ------------- |
| Dropout probability | 0.25 |
| Learning rate | 1e-1 (0.1) |
| Number of training epochs | 800 |
| Number of hidden gcn units | 2 |
| Number of hidden gcn layers | 1 |
| Pseudo coordinate dimensions in GMMConv, 2 for cora and 3 for pubmed | 2 |
| Number of kernels in GMMConv layer | 3 |
| Weight for L2 loss | 5e-4 (0.0005) |

## GAT

StatisticsHyperparameters

| Statistic | Value |
| --------- | ----- |
| Accuracy | 0.753 |
| Precision | 0.726 |
| Recall | 0.684 |
| F-Score | 0.697 |

| Hyperparameter | Value |
| ------------------------------------------ | ----------- |
| Number of training epochs | 1000 |
| Number of hidden attention heads | 4 |
| Uumber of output attention heads | 1 |
| Number of hidden layers | 1 |
| Number of hidden units | 200 |
| Use residual connection | False |
| Input feature dropout | 0 |
| Attention dropout | 0 |
| Learning rate | 1e-2 (0.01) |
| Weight decay | 0 |
| The negative slope of leaky relu | 0.2 |
| Indicates whether to use early stop or not | False |
| Skip re-evaluate the validation set | False |

## GCN

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.8024691358024691 |
| Precision | 0.8550170307357392 |
| Recall | 0.786734693877551 |
| F-Score | 0.8035058227176454 |

| Hyperparameter | Value |
| --------------------------- | ----------- |
| Dropout probability | 0 |
| Learning rate | 1e-2 (0.01) |
| Number of training epochs | 4000 |
| Number of hidden gcn units | 500 |
| Number of hidden gcn layers | 1 |
| Weight for L2 loss | 0 |

## APPNP

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.8888888888888888 |
| Precision | 0.9251082251082252 |
| Recall | 0.8943877551020407 |
| F-Score | 0.9049666689418242 |

| Hyperparameter | Value |
| --------------------------- | ------------- |
| Input feature dropout | 0.25 |
| Edge propagation dropout | 0.5 |
| Learning rate | 1e-1 (0.1) |
| Number of training epochs | 800 |
| Hidden unit sizes for appnp | [64] |
| Number of propagation steps | 10 |
| Teleport Probability | 0.4 |
| Weight for L2 loss | 5e-4 (0.0005) |

## GIN

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.8271604938271605 |
| Precision | 0.8205627705627706 |
| Recall | 0.8091836734693878 |
| F-Score | 0.8045525902668759 |

| Hyperparameter | Value |
| ------------------------- | ---------------- |
| Extra args | [16, 1, 0, True] |
| Learning rate | 1e-2 (0.01) |
| Weight decay | 5e-6 (0.000005) |
| Number of training epochs | 800 |

## TAGCN

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.9012345679012346 |
| Precision | 0.9070381998953428 |
| Recall | 0.9102607709750566 |
| F-Score | 0.9041060526774812 |

| Hyperparameter | Value |
| ------------------------- | -------------------- |
| Extra args | [16, 1, F.relu, 0.5] |
| Learning rate | 1e-2 (0.01) |
| Weight decay | 5e-4 (0.0005) |
| Number of training epochs | 800 |

## SGC

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.8271604938271605 |
| Precision | 0.8362389490209041 |
| Recall | 0.8315759637188209 |
| F-Score | 0.8240298807695121 |

| Hyperparameter | Value |
| ------------------------- | ---------------- |
| Extra args | [None, 1, False] |
| Learning rate | 1e-1 (0.1) |
| Weight decay | 0 |
| Number of training epochs | 4000 |

## AGNN

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.8765432098765432 |
| Precision | 0.88992673992674 |
| Recall | 0.8888321995464853 |
| F-Score | 0.8850179383028748 |

| Hyperparameter | Value |
| ------------------------- | ------------------------ |
| Extra args | [100, 1, 1.0, True, 0.1] |
| Learning rate | 1e-1 (0.1) |
| Weight decay | 0 |
| Number of training epochs | 200 |

## ChebNet

StatisticsHyperparameters

| Statistic | Value |
| --------- | ------------------ |
| Accuracy | 0.9012345679012346 |
| Precision | 0.9022735409953455 |
| Recall | 0.9201814058956915 |
| F-Score | 0.9073651359365645 |

| Hyperparameter | Value |
| ------------------------- | ---------------- |
| Extra args | [32, 1, 2, True] |
| Learning rate | 1e-2 (0.001) |
| Weight decay | 5e-4 (0.0005) |
| Number of training epochs | 800 |

## Feature Engineering

```sh
python generate_data.py --num_classes=6
python feature_engineering.py
```