How omarmohamed15/scalodeep works
Project Type and Purpose
• SCALODEEP is an open-source scientific tool/framework (Python + Jupyter notebooks) for deep-learning–based real-time earthquake signal detection.
• It provides data‐preprocessing utilities, sliding‐window generators, model training/inference code, pretrained weights, and evaluation scripts.
High-Level Components
a. Data Layer
• Raw seismic data (Test40Hz.mseed)
• Preprocessed/packaged NumPy archives (Data.npz, SCALO.npz)
b. Configuration
• PreTrained_SCALODEEP.json (model architecture & hyperparameters)
• LayersWeights folder (per‐layer weight files in .npy/.npz)
c. Utilities
• SCALO_Utils.py (data loading, windowing, normalization)
• Jupyter notebooks for each pipeline stage:
– Overlapping/non-overlapping window generation
– Continuous data handling & label assignment
– Model evaluation & detection logging
d. Model Runtime
• Inference engine loading JSON + weight files → produces SCALODEEP_DETECTED.txt
e. External Interfaces
• Mega.nz download link for pretrained model
• Standard file system (local read/write of weights, outputs)
Relationships & Data Flow
1. Raw seismic stream (MSEED) is ingested by SCALO_Utils → transforms into numeric time windows → saved into Data.npz / SCALO.npz.
2. Jupyter notebooks reference these .npz datasets + JSON config → instantiate model architecture → load per-layer weights from LayersWeights.
3. Trained model processes new data windows → outputs detection events → writes SCALODEEP_DETECTED.txt.
Architectural Patterns & Principles
• Pipeline pattern: clear stages (Ingest → Preprocess → Model Construct → Inference → Output).
• Separation of concerns:
• Data I/O isolated in SCALO_Utils
• Model configuration (JSON) separated from weights (LayersWeights)
• Experiment orchestration via notebooks
• Modular extensibility: new preprocessing routines or model variants can be plugged in by updating SCALO_Utils and JSON.