This is an open-source toolkit for Heterogeneous Graph Neural Network(OpenHGNN) based on DGL [Deep Graph Library] and PyTorch.

Last update: Jan 02, 2023

Overview

OpenHGNN

This is an open-source toolkit for Heterogeneous Graph Neural Network(OpenHGNN) based on DGL [Deep Graph Library] and PyTorch. We integrate SOTA models of heterogeneous graph.

Key Features

Easy-to-Use: OpenHGNN provides easy-to-use interfaces for running experiments with the given models and dataset. Besides, we also integrate optuna to get hyperparameter optimization.
Extensibility: User can define customized task/model/dataset to apply new models to new scenarios.
Efficiency: The backend dgl provides efficient APIs.

Get Started

Requirements and Installation

Python >= 3.6
PyTorch >= 1.7.1
DGL >= 0.7.0
CPU or NVIDIA GPU, Linux, Python3

1. Python environment (Optional): We recommend using Conda package manager

conda create -n openhgnn python=3.7
source activate openhgnn

2. Pytorch: Install PyTorch. For example:

# CUDA versions: cpu, cu92, cu101, cu102, cu101, cu111
pip install torch==1.8.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

3. DGL: Install DGL, follow their instructions. For example:

# CUDA versions: cpu, cu101, cu102, cu110, cu111
pip install --pre dgl-cu101 -f https://data.dgl.ai/wheels-test/repo.html

4. OpenHGNN and other dependencies:

git clone https://github.com/BUPT-GAMMA/OpenHGNN
cd OpenHGNN
pip install -r requirements.txt

Running an existing baseline model on an existing benchmark dataset

python main.py -m model_name -d dataset_name -t task_name -g 0 --use_best_config

usage: main.py [-h] [--model MODEL] [--task TASK] [--dataset DATASET] [--gpu GPU] [--use_best_config]

optional arguments: -h, --help show this help message and exit

--model MODEL, -m MODEL name of models

--task TASK, -t TASK name of task

--dataset DATASET, -d DATASET name of datasets

--gpu GPU, -g GPU controls which gpu you will use. If you do not have gpu, set -g -1.

--use_best_config use_best_config means you can use the best config in the dataset with the model. If you want to set the different hyper-parameter, modify the openhgnn.config.ini manually. The best_config will override the parameter in config.ini.

--use_hpo Besides use_best_config, we give a hyper-parameter example to search the best hyper-parameter automatically.

e.g.:

python main.py -m GTN -d imdb4GTN -t node_classification -g 0 --use_best_config

It is under development, and we release it in a nightly build version. For now, we just give some new models, such as HetGNN, NSHE, GTN, MAGNN, RSHN.

Note: If you are interested in some model, you can refer to the below models list.

Refer to the docs to get more basic and depth usage.

Models

Supported Models with specific task

The link will give some basic usage.

Model	Node classification	Link prediction	Recommendation
RGCN[ESWC 2018]	✔️	✔️
HAN[WWW 2019]	✔️
KGCN[WWW 2019]			✔️
HetGNN[KDD 2019]	✔️	✔️
GTN[NeurIPS 2019]	✔️
RSHN[ICDM 2019]	✔️
DGMI[AAAI 2020]	✔️
MAGNN[WWW 2020]	✔️
CompGCN[ICLR 2020]	✔️	✔️
NSHE[IJCAI 2020]	✔️
NARS[arxiv]	✔️
MHNF[arxiv]	✔️
HGSL[AAAI 2021]	✔️
HGNN-AC[WWW 2021]	✔️
HPN[TKDE 2021]	✔️
RHGNN[arxiv]	✔️

To be supported models

Metapath2vec[KDD 2017]

Candidate models

Heterogeneous Graph Attention Networks for Semi-supervised Short Text Classification[EMNLP 2019]
Heterogeneous Graph Structure Learning for Graph Neural Networks[AAAI 2021]
Self-supervised Heterogeneous Graph Neural Network with Co-contrastive Learning[KDD 2021]
Heterogeneous Information Network Embedding with Adversarial Disentangler[TKDE 2021]

Contributors

GAMMA LAB [BUPT]: Tianyu Zhao, Yaoqi Liu, Fengqi Liang, Yibo Li, Yanhu Mo, Donglin Xia, Xinlong Zhai, Siyuan Zhang, Qi Zhang, Chuan Shi, Cheng Yang, Xiao Wang

BUPT: Jiahang Li, Anke Hu

DGL Team: Quan Gan, Jian Zhang

Comments

Attribute error

I am training HetGNN model for node classification. when i try to run the script for training. I get the following error. Please help me AttributeError: 'dict' object has no attribute 'srcdata'

opened by faizan1234567 13
error in HetGNN_sampler.py

line 168, in assign_features_to_blocks assign_simple_node_features(blocks[0].srcdata, g, ntypes) AttributeError: 'dict' object has no attribute 'srcdata'

opened by Kingrd97 10
关于HetGNN-emb有完全相同的情况

通过HetGNN跑提供的academic4HetGNN.zip 数据集。emb结果有完全相同的情况发生，原因未知。请问是否是符合预期的？

如下测试：

`import numpy as np

emb = np.load('emb50.npy') list = emb[:,0]

for i in np.unique(list): idx = np.argwhere(list == i) r = idx.reshape(1, -1).squeeze(0) if len(r) > 1: print('index for {}:\n'.format(i), r) for j in r: print(emb[j]) `

opened by lixusign 9
Error to run without Cuda

File "C:\Users\XyZ\OpenHGNN\openhgnn\models\GTN_sparse.py", line 220, in forward sum_g = dgl.adj_sum_graph(A, 'w_sum') AttributeError: module 'dgl' has no attribute 'adj_sum_graph'

This Issue came-up while I ran the command= python main.py -m GTN -d imdb4GTN -t node_classification -g -1 --use_best_config

Can someone say me where I went wrong?

opened by M-Somtirth 4
无法使用gpu训练

python main.py -m KGCN -d LastFM4KGCN -t recommendation -g 0 --use_best_config

RuntimeError: Tensor for argument #2 'mat1' is on CPU, but expected it to be on GPU (while checking arguments for addmm)

opened by Tingting-Liu-star 4
where I can find the dataset build program ?

for ex : https://github.com/BUPT-GAMMA/OpenHGNN/tree/main/openhgnn/dataset#academic4HetGNN in this dataset

when I extract_archive , A bin file about g。

but where I can find ， how to build this dataset use dgl standalone program ?

opened by lixusign 4
Error when running GTN&fastGTN

Thank you very much for being able to provide this tool. I get an error when I run fastGTN using:

python main.py -m fastGTN -t node_classification -d acm4GTN -g 0 --use_best_config

The error is as follows：

Traceback (most recent call last): File "D:/github/OpenHGNN/main.py", line 30, in OpenHGNN(args=config) File "D:\github\OpenHGNN\openhgnn\start.py", line 19, in OpenHGNN result = flow.train() File "D:\github\OpenHGNN\openhgnn\trainerflow\node_classification.py", line 112, in train train_loss = self._full_train_step() File "D:\github\OpenHGNN\openhgnn\trainerflow\node_classification.py", line 152, in _full_train_step logits = self.model(self.hg, h_dict)[self.category] File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "D:\github\OpenHGNN\openhgnn\models\fastGTN.py", line 119, in forward hat_A = self.layersi File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "D:\github\OpenHGNN\openhgnn\models\fastGTN.py", line 180, in forward sum_g = dgl.adj_sum_graph(A, 'w_sum') File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl\transforms\functional.py", line 2766, in adj_sum_graph C_gidx, C_weights = F.csrsum(gidxs, weights) File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl\backend\pytorch\sparse.py", line 817, in csrsum nrows, ncols, C_indptr, C_indices, C_eids, C_weights = CSRSum.apply(gidxs, *weights) File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl\backend\pytorch\sparse.py", line 668, in forward gidxC, C_weights = _csrsum(gidxs, weights) File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl\sparse.py", line 776, in _csrsum C, C_weights = _CAPI_DGLCSRSum(As, [F.to_dgl_nd(w) for w in A_weights]) File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl_ffi_ctypes\function.py", line 188, in call check_call(_LIB.DGLFuncCall( File "D:\Program Files (x86)\anaconda\envs\OpenHGNN\lib\site-packages\dgl_ffi\base.py", line 65, in check_call raise DGLError(py_str(_LIB.DGLGetLastError())) dgl._ffi.base.DGLError: [15:31:21] C:\Users\Administrator\dgl-0.5\src\array\kernel.cc:471: Check failed: A[i].indptr->dtype == idtype (int64 vs. int32) : The ID types of all graphs must be equal.

I use the following software versions:

python = 3.8 cudatoolkit = 11.3.1 torch = 1.11.0+cu113 dgl-cu113 = 0.8.1 & 0.8.0

Then I ran the same version of the software on my ubuntu server with no errors.

opened by huihuijiangqiang 3

bugs in minibatch trainning

🐛 Bug

To Reproduce

error occurred in _mini_train_step function in trainerflow/node_classification.py when use mini_batch_flag in node_classification task and SimpleHGN model

import argparse
from openhgnn.experiment import Experiment

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--model', '-m', default='SimpleHGN', type=str, help='name of models')
    parser.add_argument('--task', '-t', default='node_classification', type=str, help='name of task')
    # link_prediction / node_classification
    parser.add_argument('--dataset', '-d', default='imdb4MAGNN', type=str, help='name of datasets')
    parser.add_argument('--gpu', '-g', default='0', type=int, help='-1 means cpu')
    parser.add_argument('--use_best_config', action='store_true', help='will load utils.best_config')
    parser.add_argument('--load_from_pretrained', action='store_true', help='load model from the checkpoint')
    args = parser.parse_args()

    experiment = Experiment(model=args.model, dataset=args.dataset, task=args.task, gpu=args.gpu,
                            use_best_config=args.use_best_config, load_from_pretrained=args.load_from_pretrained, mini_batch_flag = True, batch_size=64)
    experiment.run()

Expected behavior

Minibatch training on a large heterograph

Environment

torch==1.12.1
dgl-cu113==0.9.0 # for CUDA support
openhgnn==0.3.0
Linux
Python 3.8.13

Additional context

the default minibatch sampler is MultiLayerFullNeighborSampler
the blocks is a list (line 164) and the expected input in the forward function of the model (e.g. SimpleHGN) is a hg(line 159)

for i, (input_nodes, seeds, blocks) in enumerate(loader_tqdm):
    blocks = [blk.to(self.device) for blk in blocks]
    ...
    logits = self.model(blocks, emb)[self.category]

def forward(self, hg, h_dict):
    with hg.local_scope():
        hg.ndata['h'] = h_dict

opened by suxnju 2

关于HetGNN的emb顺序困惑

请教下在 x = self.model(blocks[0], input_features) 中返回的x 是dict 。他里面每种node_type 的emb 和blocks[0] 的入参的点的顺序如何对应？

我核对了以后发现并不是 blocks[0].srcnodes[node_type].data[dgl.NID] 所代表的节点顺序。

opened by lixusign 2
HIN_LinkPrediction' object has no attribute 'get_idx'

\OpenHGNN-main\openhgnn\tasks\link_prediction.py", line 32, in init self.train_hg, self.val_hg, self.test_hg = self.dataset.get_idx() AttributeError: 'HIN_LinkPrediction' object has no attribute 'get_idx'

opened by xuptacm 2
无法使用ACM4GTN数据集运行GTN

运行 python main.py -m GTN -t node_classification -d acm4GTN -g 0 --use_best_config

报错信息 Using backend: pytorch Use the best config. Done saving data into cached files. Modify the out_dim with num_classes 0%| | 0/50 [00:00<?, ?it/s] Traceback (most recent call last): File "main.py", line 24, in OpenHGNN(args=config) File "/home/special/user/lihaoran/OpenHGNN_clone_from_github/openhgnn/start.py", line 17, in OpenHGNN result = flow.train() File "/home/special/user/lihaoran/OpenHGNN_clone_from_github/openhgnn/trainerflow/node_classification.py", line 77, in train loss = self._full_train_step() File "/home/special/user/lihaoran/OpenHGNN_clone_from_github/openhgnn/trainerflow/node_classification.py", line 109, in _full_train_step loss.backward() File "/opt/miniconda3/lib/python3.7/site-packages/torch/_tensor.py", line 255, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/opt/miniconda3/lib/python3.7/site-packages/torch/autograd/init.py", line 149, in backward allow_unreachable=True, accumulate_grad=True) # allow_unreachable flag File "/opt/miniconda3/lib/python3.7/site-packages/torch/autograd/function.py", line 87, in apply return self._forward_cls.backward(self, *args) # type: ignore[attr-defined] File "/opt/miniconda3/lib/python3.7/site-packages/dgl/backend/pytorch/sparse.py", line 544, in backward gidxA.reverse(), A_weights, gidxC, dC_weights, gidxB.number_of_ntypes()) File "/opt/miniconda3/lib/python3.7/site-packages/dgl/backend/pytorch/sparse.py", line 638, in csrmm CSRMM.apply(gidxA, A_weights, gidxB, B_weights, num_vtypes) File "/opt/miniconda3/lib/python3.7/site-packages/dgl/backend/pytorch/sparse.py", line 528, in forward gidxC, C_weights = _csrmm(gidxA, A_weights, gidxB, B_weights, num_vtypes) File "/opt/miniconda3/lib/python3.7/site-packages/dgl/sparse.py", line 548, in _csrmm A, F.to_dgl_nd(A_weights), B, F.to_dgl_nd(B_weights), num_vtypes) File "dgl/_ffi/_cython/./function.pxi", line 287, in dgl._ffi._cy3.core.FunctionBase.call File "dgl/_ffi/_cython/./function.pxi", line 232, in dgl._ffi._cy3.core.FuncCall File "dgl/_ffi/_cython/./base.pxi", line 155, in dgl._ffi._cy3.core.CALL dgl._ffi.base.DGLError: [17:18:53] /opt/dgl/src/array/cuda/csr_mm.cu:87: Check failed: e == CUSPARSE_STATUS_SUCCESS: CUSPARSE ERROR: 11 Stack trace: [bt] (0) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x4f) [0x7fd13c2565df] [bt] (1) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(std::pair<dgl::aten::CSRMatrix, dgl::runtime::NDArray> dgl::aten::cusparse::CusparseSpgemm<float, int>(dgl::aten::CSRMatrix const&, dgl::runtime::NDArray, dgl::aten::CSRMatrix const&, dgl::runtime::NDArray)+0x625) [0x7fd13c6accd5] [bt] (2) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(std::pair<dgl::aten::CSRMatrix, dgl::runtime::NDArray> dgl::aten::CSRMM<2, long, float>(dgl::aten::CSRMatrix const&, dgl::runtime::NDArray, dgl::aten::CSRMatrix const&, dgl::runtime::NDArray)+0x59e) [0x7fd13c6af81e] [bt] (3) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(dgl::aten::CSRMM(dgl::aten::CSRMatrix, dgl::runtime::NDArray, dgl::aten::CSRMatrix, dgl::runtime::NDArray)+0x10d6) [0x7fd13c493466] [bt] (4) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(+0x48cfa8) [0x7fd13c493fa8] [bt] (5) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(+0x48d724) [0x7fd13c494724] [bt] (6) /opt/miniconda3/lib/python3.7/site-packages/dgl/libdgl.so(DGLFuncCall+0x48) [0x7fd13c4d5c78] [bt] (7) /opt/miniconda3/lib/python3.7/site-packages/dgl/_ffi/_cy3/core.cpython-37m-x86_64-linux-gnu.so(+0x163ea) [0x7fd1136f03ea] [bt] (8) /opt/miniconda3/lib/python3.7/site-packages/dgl/_ffi/_cy3/core.cpython-37m-x86_64-linux-gnu.so(+0x1695b) [0x7fd1136f095b]

GPU：A100-PCIE DGL版本：dgl-cu111-0.8a211008 看起来可以获得logits，但是无法进行反向传播

很奇怪的是运行IMDB4GTN数据集时没有任何问题使用MHNF运行ACM4GTN也报了同样的错误

我看到GTN有两个，一个是叫GTN_spare.py一个GTN.py，默认是用的GTN_spare。用GTN.py可以运行ACM4GTN，但是准确率只有60%上下

opened by a772316182 2
Help needed: Wanted behavior of Experiment.specific_trainerflow.get method and task/trainerflow registration

Hi, I am trying to create a new trainer flow, as well as a new task. I am struggling a bit and have a few questions: When I register them with @register_flow(str_flow) and @register_task(str_task), must str_taskand str_flowbe identical?
Because as my flow is not specific to a model, it is not in the specific_trainerflowdictionnary defined in the Experiment class. So the line 92 in experiment.py( trainerflow = self.specific_trainerflow.get(self.config.model, self.config.task) ) returns the key of the task as the trainerflow_key. Is this the wanted behavior?

Thanks!

opened by Carayolj 0
run HGSL model error
🐛 Bug

when i run the suggest command :

python main.py -m HGSL -d acm4GTN -t node_classification -g 0 --use_best_config

this raise an error like:

Traceback (most recent call last): File "main.py", line 21, in experiment.run() File "/workspace/OpenHGNN/openhgnn/experiment.py", line 97, in run flow = build_flow(self.config, trainerflow) File "/workspace/OpenHGNN/openhgnn/trainerflow/init.py", line 46, in build_flow return FLOW_REGISTRYflow_name File "/workspace/OpenHGNN/openhgnn/trainerflow/node_classification.py", line 42, in init self.model = build_model(self.model).build_model_from_args(self.args, self.hg).to(self.device) File "/workspace/OpenHGNN/openhgnn/models/HGSL.py", line 106, in build_model_from_args mp_emb_dim = hg.nodes["paper"].data["pap_m2v_emb"].shape[1] File "/opt/conda/lib/python3.7/site-packages/dgl/view.py", line 73, in getitem return self._graph._get_n_repr(self._ntid, self._nodes)[key] File "/opt/conda/lib/python3.7/site-packages/dgl/frame.py", line 622, in getitem return self._columns[name].data KeyError: 'pap_m2v_emb'

it seems like there is no pap_m2v_emb key in paper nodes data， so how to fix it?

more error update: when I just make mp_emb_dim=0 to jump this line, more errors raise, such as no hidden_dim、mini_batch_flag ... defined in config, besides, when I successfully run this model, another exception was raised:

Do you have an updated version of the model？

Sincere thanks.

To Reproduce

Steps to reproduce the behavior:

1.cd OpenHGNN 2.python main.py -m HGSL -d acm4GTN -t node_classification -g 0 --use_best_config

Expected behavior

Environment

OpenHGNN Version (e.g., 1.0):

PyTorch latest, DGL latest

Linux

python main.py -m HGSL -d acm4GTN -t node_classification -g 0 --use_best_config

best_config for recommend
opened by vchopin 1
How to train model using own dataset？

❓ Questions and Help

I want to train my own HNN data, could you tell me how to edit this code? the data in ./openhgnn/dataset are download from https://s3.cn-north-1.amazonaws.com.cn/dgl-data/ and is .bin file. So how could I change this dataset? 救救孩子

opened by Fino2020 1
[DHNE]
Description

Checklist

Please feel free to remove inapplicable items for your PR.

[x] The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])

[x] Changes are complete (i.e. I finished coding on this PR)

[x] All changes have test coverage

[x] Code is well-documented

[x] To the best of my knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

[x] Related issue is referred in this PR

[x] If the PR is for a new model/paper, I've updated the example index here.

Changes
opened by Vera-200 0
[Model]Mg2vec
Description

Add the Mg2vec Model and add the EdgeClassification Task

Checklist

Please feel free to remove inapplicable items for your PR.

[ ] The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])

[ ] Changes are complete (i.e. I finished coding on this PR)

[ ] All changes have test coverage

[ ] Code is well-documented

[ ] To the best of my knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

[ ] Related issue is referred in this PR

[ ] If the PR is for a new model/paper, I've updated the example index here.

Changes

[ ] Add configs for Mg2vec in config.ini and config.py

[ ] Add Mg2vec.py, which contains the model part

[ ] Add mg2vec_sampler.py for reading data

[ ] Add mg2vec_trainer.py for training

[ ] Add EdgeClassificationDataset.py for EdgeClassification Task, which is a modified version of NodeClassificationDataset.py

[ ] Add mg2vec_dataset.py for download/read mg2vec dataset

[ ] Add edge_classification.py, which is a modified version of node_classification.py

[ ] Add ec_with_SVC function in evaluator.py for edge_classification task

[ ] Add readme.md for Mg2vec model

[ ] Modify corresponding init.py and experiment.py
opened by null-xyj 0

Releases(v0.3.0)

v0.3.0(Jun 26, 2022)
New models

API usage

Customization of datasets and models

Source code(tar.gz)
Source code(zip)
v0.2.0(Mar 1, 2022)
New models

Space4HGNN

Benchmark & Leaderboard

Source code(tar.gz)
Source code(zip)
v0.1.1(Jan 7, 2022)
Release v0.1.1 on OpenI with Chinese.

Add more models.

Add logger feature.

More models implemented in link prediction

Source code(tar.gz)
Source code(zip)
v0.1.0(Sep 14, 2021)

The first open release includes basically everything in the repository.

Docs: a tutorial for basic and depth usage. Models: 16 HGNN models
Source code(tar.gz)
Source code(zip)

Owner

BUPT GAMMA Lab

Graph dAta Mining and MAchine learning Lab at Beijing University of Posts and Telecommunications

GitHub Repository

This is an open-source toolkit for Heterogeneous Graph Neural Network(OpenHGNN) based on DGL [Deep Graph Library] and PyTorch.

Related tags

Overview

OpenHGNN

Key Features

Get Started

Requirements and Installation

Running an existing baseline model on an existing benchmark dataset

Supported Models with specific task

To be supported models

Candidate models

Contributors

Comments

🐛 Bug

To Reproduce

Expected behavior

Environment

Additional context

🐛 Bug

To Reproduce

Expected behavior

Environment

❓ Questions and Help

Description

Checklist

Changes

Description

Checklist

Changes

Releases(v0.3.0)

v0.3.0(Jun 26, 2022)

v0.2.0(Mar 1, 2022)

v0.1.1(Jan 7, 2022)

v0.1.0(Sep 14, 2021)

Owner

BUPT GAMMA Lab

Library for time-series-forecasting-as-a-service.

DockStream: A Docking Wrapper to Enhance De Novo Molecular Design

Plaything for Autistic Children (demo for PaddlePaddle/Wechaty/Mixlab project)

Official Keras Implementation for UNet++ in IEEE Transactions on Medical Imaging and DLMIA 2018

Deep Learning Datasets Maker is a QGIS plugin to make datasets creation easier for raster and vector data.

A PyTorch Implementation of "Watch Your Step: Learning Node Embeddings via Graph Attention" (NeurIPS 2018).

An SE(3)-invariant autoencoder for generating the periodic structure of materials

Stacked Recurrent Hourglass Network for Stereo Matching

QTool: A Low-bit Quantization Toolbox for Deep Neural Networks in Computer Vision

ViViT: Curvature access through the generalized Gauss-Newton's low-rank structure

The codes I made while I practiced various TensorFlow examples

Localizing Visual Sounds the Hard Way

A containerized REST API around OpenAI's CLIP model.

Hands-On Machine Learning for Algorithmic Trading, published by Packt

Deep Learning Slide Captcha

TorchX: A PyTorch Extension Library for More Efficient Deep Learning

Simple codebase for flexible neural net training

Transformers are Graph Neural Networks!

Syntax-Aware Action Targeting for Video Captioning

QHack—the quantum machine learning hackathon