Financial Anomaly Detection Task

Dataset

• Dataset name: DGraph
• Dataset source: DGraph
• Dataset description: A collection of large‑scale dynamic graph datasets built from real financial scenarios where events and labels evolve over time.

Experiment Objectives

• Learn how to train neural networks using PyTorch
• Learn how to design simple graph neural networks with libraries such as PyTorch‑Geometric (recommended models: GAT, GraphSAGE)
• Learn how to use the MO platform for model performance evaluation.

Prerequisite Knowledge

• Basic deep learning theory including CNNs, loss functions, optimizers, training strategies, etc.
• Familiarity with the PyTorch framework.
• To learn PyTorch‑Geometric, visit: https://pytorch-geometric.readthedocs.io/en/latest/

Experiment Environment

• python = 3.9.5
• numpy = 1.26.4
• pandas =
• pytorch = 2.3.1
• torch_geometric = 2.5.3
• torch_scatter = 2.1.2
• torch_sparse = 0.6.18

Environment Setup

1. Install Miniconda or Anaconda; see the official documentation: https://www.anaconda.com/
2. Create a conda environment

   conda create -n mo_graph python=3.9.5
   conda activate mo_graph
   

3. Install PyTorch following the official guide: https://pytorch.org/get-started/previous-versions/

   conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==0.13.1 pytorch-cuda=11.8 -c pytorch -c nvidia
   conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==0.13.1 pytorch-cuda=12.1 -c pytorch -c nvidia
   conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==0.13.1 cpuonly -c pytorch
   

4. Install PyG related packages, including torch_geometric, torch_sparse, torch_scatter: https://github.com/pyg-team/pytorch_geometric

   pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.3.1+${CUDA}.html
   pip install torch_geometric==2.5.3
   

5. Install numpy and other dependencies

   conda install numpy=1.26.4 pandas