dnn-modeling

What problem does it solve?

This Skill automates deep learning model training for binary classification and produces a complete, finance-ready evaluation report (AUC/KS/BCR/Lift and stability analysis) so you don’t have to assemble the workflow manually.

Core Features & Use Cases

• MLP (DNN) Binary Modeling with PyTorch: trains a multi-layer perceptron for 0/1 classification using BatchNorm, Dropout, learning-rate scheduling, and early stopping.
• Three-way dataset evaluation: performs standardized splits into Train/Val/OOT and evaluates generalization using consistent AUC/KS/BCR/Lift reporting.
• Robust preprocessing & safety checks: reuses platform splitting, missing-value imputation (median + missing indicators), feature standardization, and a pre-training SafetyGate to detect time leakage and schema issues.
• Use Case: You have a parquet/csv dataset with high-dimensional features and a 0/1 target; you want a production-style model card and performance diagnostics across OOT to compare against XGBoost and LR.

Quick Start

Train an MLP binary model on your dataset by asking the AI: "Use the dnn-modeling skill with --data_path ./data.parquet --target y_label --time_col busi_dt and write the outputs to ./outputs/dnn_run."