> This page is for version 1.0.0. > For other versions, use one of these documentation indexes: > - 3.0.0 (Latest) (default): https://docs.nvidia.com/nim/financial-fraud-training/3.0.0/llms.txt > - 2.0.0: https://docs.nvidia.com/nim/financial-fraud-training/2.0.0/llms.txt > - 1.0.0: https://docs.nvidia.com/nim/financial-fraud-training/1.0.0/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/nim/financial-fraud-training/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nim/financial-fraud-training/_mcp/server. # Graph Data Organization for GNN-based XGBoost Training Organize your data under a parent directory (for example, `data_root`) with the following subdirectories: ```sh data_root/ ├── nodes/ └── edges/ ``` - **nodes/**: Contains the files for node features and labels. - **edges/**: Contains the file for edges. --- ## Nodes Directory ### File Naming Convention - **Node Feature File:** Name the file as `node.`, where `` is one of the supported file formats [(CSV, Parquet, or ORC)](https://en.wikipedia.org/wiki/Data_orientation). Example: - `node.csv` - **Node Label File:** Provide the labels for your nodes in a file named `node_label.`. Example: - `node_label.csv` - **Optional offset range JSON file:** Provide the start and end offset of the target nodes in the `node.` file. If this isn't provided, the training will be run on all the nodes. - `offset_range_of_training_node.json` ### File Contents - **Feature File (`node.`):** This file should contain a table of node attributes (features). - Each row corresponds to one node. - The number of columns is the feature dimension. NOTE: If you have nodes of different types, You must prepare and align their feature.ach row must contain the same number of feature entries. - **Label File (`node_label.`, Optional):** This file should contain the ground-truth labels for the nodes. - There must be one column for each name, with the number of rows matching the feature file. - **Optional** JSON file containing the starting and ending offsets for the nodes in the `node.` file that will be used for training and validation. In other words, it defines the range of node offsets that the model uses to train. Consider credit card fraud detection. A `node.` file includes features for Users, Merchants, and Transaction nodes. Your goal is to find fraudulent transactions. Here, you can only train on transaction nodes. ```sh { "start": ST #start offset of training nodes "end": ET #end offset of training nodes } ``` If no file is provided, the models will be trained using the entire range. --- ## Edges Directory ### File Naming Convention - **Edge File:** Name the file as `node_to_node.`, where `` is one of the supported file formats (CSV, Parquet, or ORC). Examples: - `node_to_node.csv` #### File Contents - **Edge File (`node_to_node.`):** This file must contain at least two columns: - `src`: The column containing indices of the source nodes - `dst`: The column containing indices of the destination nodes. - The index of a node is determined by its position in the node feature file `node.` and is zero-based. - The number of rows corresponds to the number of edges. NOTE: The columns containing source and destination nodes must be named `src` and `dst`, respectively. --- ### Data types #### `node.` - **Description:** Contains node features. - **Data Type:** All columns must contain floating point numbers in `float32` format. - **Notes:** Each row represents a node, and each column corresponds to a specific feature. #### `node_label.` - **Description:** Contains node labels. - **Data Type:** All values are integers (0 or 1). - **Column Name:** The file contains a single column named `target`. - **Notes:** Each row corresponds to the label of a node. #### `node_to_node.` - **Description:** Contains edge indices. - **Data Type:** All columns must contain integers. - **Columns:** - `src`: Source node index. - `dst`: Destination node index. - **Notes:** The indices must be zero-based; in other words, node numbering starts at 0. ### Example Data Layout ```sh credit_card_transaction_data/ ├── nodes/ │ ├── node.csv │ └── node_label.csv │ └── offset_range_of_training_node.json └── edges/ └── node_to_node.csv ``` ### Data Organization for Standalone XGBoost Training Your dataset is expected to be organized in a specific structure under the `xgb/` folder. Follow the guidelines below to ensure your data is correctly formatted and named. ### Directory Structure Place your data files under the `xgb/` directory within your main data folder. The expected files are: - **`training.`** (Required): Contains the training data. - **`validation.`** (Optional): Contains the validation data. - If this file is missing, the training data will be split into 80% training and 20% validation. - **`test.`** (Optional): Contains the test data. If test data is provided, it will be used for testing the model performance. ### Supported File Formats The `` in the file names represents the file format. The supported file formats are: - **CSV**: Files with the `.csv` extension (comma-separated values). - **Parquet**: Files with the `.parquet` extension (columnar storage). - **ORC**: Files with the `.orc` extension (Optimized Row Columnar format). #### Important Data Format Note - **Target Column:** The last column in every file (`i.e.training.`, `validation.` and `test.`) is assumed to be the target variable (y). All splits (training, validation, and test) are performed with this assumption in mind. #### Example Directory Layout ```sh data_root/ └── xgb/ ├── training.csv # or training.parquet, training.orc ├── validation.csv # (optional) or validation.parquet, validation.orc └── test.csv # (optional) or test.parquet, test.orc ``` #### Data Splitting Behavior - **When only `training.` is provided:** - If **only the `training.` file exists**, the code will split the data into 80% training and 20% validation. - **When `validation.` is provided:** - The provided training and validation files are used directly without splitting. - **When `test.` is present:** - The test file is used to evaluate the model after training. #### Model Testing If a `test.` file is present in the `xgb/` directory, it will be used for evaluation after training.