How coderdotexe/hbpa works
Project Type and Purpose
• This is a Python-based machine learning pipeline for handwriting-based psychological assessment.
• The system ingests scanned handwriting images, applies preprocessing, extracts features, labels data, trains/predicts via ML models (SVM by default), and outputs trait predictions.
Main Architectural Components
a. Data Layer
• Source: IAM Handwriting Database (external).
• Local storage:
• images/ (processed PNGs)
• data/samples/ (raw images, per README)
• feature_list, raw_feature_list, label_list (flat files storing intermediate data).
b. Ingestion & Preprocessing
• Scripts: extract.py, extract_v3.py, extract_routine.py
• Tasks: resolution enhancement, noise reduction, grayscale conversion, contour detection, warp affinities, projections
c. Feature Extraction
• Scripts: ExtractFeature.py, feature_routine.py
• Outputs: baseline, margins, letter size, spacing metrics, pen pressure, slant angle
d. Labeling
• Scripts: label_routine.py, categorize.py
• Creates mappings from features to psychological traits (eight traits)
e. Model Training & Prediction
• Scripts: train_predict.py, train_predict_v3.py
• Libraries: scikit-learn (SVM), numpy, pandas
• Produces terminal output with predicted trait scores
f. Reporting/Output
• Terminal prints
• (Future scope: UI layer, possibly web or desktop)
Interactions and Data Flow
• External Dataset → Data Layer (raw images)
↓
• Ingestion & Preprocessing → cleans and standardizes images → images/
↓
• Feature Extraction → reads processed images, writes feature_list/raw_feature_list
↓
• Labeling → reads features, applies label_list, writes training labels
↓
• Model Training & Prediction → reads features + labels, fits/predicts → prints results
↓
• Output → Terminal (and logs if extended)
Architectural Patterns & Principles
• Pipeline Pattern: clear sequential stages
• Modular Separation: each stage in its own script/routine
• Single Responsibility: one script per processing step
• File-based messaging: flat files as intermediate queues
• Extensibility: future CNN integration, UI layer
Technologies & Libraries
• Python 3.x, OpenCV (cv2), scikit-learn (SVM), numpy, pandas
• Flat files (lists) for simple persistence
• Parent README mentions future use of CNNs
Detail Level
• .py scripts as distinct processing nodes.
• Intermediate artifact files as data stores.
• Do not merge stages—better to show more granular steps.