NVIDIA 文档中心 NVIDIA Morpheus NVIDIA Morpheus (24.10.01) GNN 欺诈检测管道

GNN 欺诈检测管道

支持的环境

所有环境都需要额外的 Conda 包，这些包可以使用 conda/environments/all_cuda-125_arch-x86_64.yaml 或 conda/environments/examples_cuda-125_arch-x86_64.yaml 环境变量文件安装。有关更多信息，请参阅要求部分。

环境	支持	注释
Conda	✔
Morpheus Docker 容器	✔
Morpheus 发布容器	✔
Dev 容器	✔

要求

在运行 GNN 欺诈检测管道之前，必须将其他要求安装到您的 Conda 环境中。

复制
已复制！

            
            conda env update --solver=libmamba \
  -n ${CONDA_DEFAULT_ENV} \
  --file ./conda/environments/examples_cuda-125_arch-x86_64.yaml

运行

使用 Morpheus 运行带有交易数据的 GNN 欺诈检测管道。管道已在 run.py 中配置，并带有多个命令行选项

复制
已复制！

            
            python examples/gnn_fraud_detection_pipeline/run.py --help

复制
已复制！

            
            Usage: run.py [OPTIONS]

Options:
  --num_threads INTEGER RANGE     Number of internal pipeline threads to use.
                                  [x>=1]
  --pipeline_batch_size INTEGER RANGE
                                  Internal batch size for the pipeline. Can be
                                  much larger than the model batch size. Also
                                  used for Kafka consumers.  [x>=1]
  --model_max_batch_size INTEGER RANGE
                                  Max batch size to use for the model.  [x>=1]
  --model_fea_length INTEGER RANGE
                                  Features length to use for the model.
                                  [x>=1]
  --input_file PATH               Input data filepath.  [required]
  --training_file PATH            Training data filepath.  [required]
  --model_dir PATH                Trained model directory path  [required]
  --output_file TEXT              The path to the file where the inference
                                  output will be saved.
  --help                          Show this message and exit.

要启动配置的 Morpheus 管道，请运行以下命令

复制
已复制！

            
            python examples/gnn_fraud_detection_pipeline/run.py

复制
已复制！

            
            ====Registering Pipeline====
====Building Pipeline====
====Building Pipeline Complete!====
====Registering Pipeline Complete!====
====Starting Pipeline====
====Pipeline Started====
====Building Segment: linear_segment_0====
Added source: <from-file-0; FileSourceStage(filename=/examples/gnn_fraud_detection_pipeline/validation.csv, iterative=False, file_type=FileTypes.Auto, repeat=1, filter_null=False, filter_null_columns=None, parser_kwargs=None)>
  └─> morpheus.MessageMeta
Added stage: <deserialize-1; DeserializeStage(ensure_sliceable_index=True, task_type=None, task_payload=None)>
  └─ morpheus.MessageMeta -> morpheus.ControlMessage
Added stage: <fraud-graph-construction-2; FraudGraphConstructionStage(training_file=/examples/gnn_fraud_detection_pipeline/training.csv)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <monitor-3; MonitorStage(description=Graph construction rate, smoothing=0.05, unit=messages, delayed_start=False, determine_count_fn=None, log_level=LogLevels.INFO)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <gnn-fraud-sage-4; GraphSAGEStage(model_dir=/examples/gnn_fraud_detection_pipeline/model, batch_size=100, record_id=index, target_node=transaction)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <monitor-5; MonitorStage(description=Inference rate, smoothing=0.05, unit=messages, delayed_start=False, determine_count_fn=None, log_level=LogLevels.INFO)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <gnn-fraud-classification-6; ClassificationStage(model_xgb_file=/examples/gnn_fraud_detection_pipeline/model/xgb.pt)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <monitor-7; MonitorStage(description=Add classification rate, smoothing=0.05, unit=messages, delayed_start=False, determine_count_fn=None, log_level=LogLevels.INFO)>
  └─ morpheus.ControlMessage -> morpheus.ControlMessage
Added stage: <serialize-8; SerializeStage(include=None, exclude=None, fixed_columns=True)>
  └─ morpheus.ControlMessage -> morpheus.MessageMeta
Added stage: <monitor-9; MonitorStage(description=Serialize rate, smoothing=0.05, unit=messages, delayed_start=False, determine_count_fn=None, log_level=LogLevels.INFO)>
  └─ morpheus.MessageMeta -> morpheus.MessageMeta
Added stage: <to-file-10; WriteToFileStage(filename=output.csv, overwrite=True, file_type=FileTypes.Auto, include_index_col=True, flush=False)>
  └─ morpheus.MessageMeta -> morpheus.MessageMeta
====Building Segment Complete!====
Graph construction rate[Complete]: 265 messages [00:00, 1016.18 messages/s]
Inference rate[Complete]: 265 messages [00:00, 545.08 messages/s]
Add classification rate[Complete]: 265 messages [00:00, 492.11 messages/s]
Serialize rate[Complete]: 265 messages [00:00, 480.77 messages/s]
====Pipeline Complete====

CLI 示例

上面的示例说明了如何使用 Python API 构建自定义 Morpheus 管道。或者，也可以使用 Morpheus 命令行来完成相同的目标。为此，我们必须确保 examples 目录在 PYTHONPATH 中可用，并且每个自定义阶段都注册为插件。

注意：由于 gnn_fraud_detection_pipeline 模块对 Python 可见，我们可以通过模块名称而不是更详细的文件路径来指定插件。

从 Morpheus 仓库的根目录运行

复制
已复制！

            
            PYTHONPATH="examples" \
morpheus --log_level INFO \
	--plugin "gnn_fraud_detection_pipeline" \
	run --pipeline_batch_size 1024 --model_max_batch_size 32 --edge_buffer_size 4 \
	pipeline-other --model_fea_length 70 --label=probs \
	from-file --filename examples/gnn_fraud_detection_pipeline/validation.csv --filter_null False \
	deserialize \
	fraud-graph-construction --training_file examples/gnn_fraud_detection_pipeline/training.csv \
	monitor --description "Graph construction rate" \
	gnn-fraud-sage --model_dir  examples/gnn_fraud_detection_pipeline/model/ \
	monitor --description "Inference rate" \
	gnn-fraud-classification --model_xgb_file examples/gnn_fraud_detection_pipeline/model/xgb.pt \
	monitor --description "Add classification rate" \
	serialize \
	to-file --filename "output.csv" --overwrite

上一个 DOCA GPU 实时流量分析

下一个 LLM