How matteoprata/lobcast works
LOBCAST is a framework for benchmarking deep-learning models that predict stock-price trends from limit order book data. The main workflow configures a simulation, loads FI-2010 samples, selects a model, trains and evaluates it, then writes metrics and PDF reports. Separate capabilities support hyperparameter sweeps, sequential experiment plans, and optional Weights & Biases logging. The graph covers the documented model families and core learning, data, and reporting layers; relationships for unsampled modules are limited to README statements and SOURCE INDEX references.
Main entry points: lobcast.py Simulation · models_classes.py Model registry · util_training.py Learning engine · lobcast_model.py Model interface
Components 27
Simulation control
- CLI entry
src/run.py - Simulation
src/lobcast.py - Settings
src/settings.py - Hyperparameters
src/hyper_parameters.py - Shared utilities
src/utils/utils_generic.py
Dataset preparation
- Dataset selection
src/data_preprocessing/utils_dataset.py - Data module
src/data_preprocessing/dataModule.py - FI-2010 samples
src/data_preprocessing/FI/FIDataBuilder.py
Prediction models
- Model selection
src/models/utils_models.py - Model registry
src/models/models_classes.py - Model interface
src/models/lobcast_model.py - MLP
src/models/mlp/mlp.py - CNN1
src/models/cnn1/cnn1.py - CNN2
src/models/cnn2/cnn2.py - BINCTABL
src/models/binctabl/binctabl.py - BINCTABL layers
src/models/binctabl/base.py
Training and evaluation
- Learning engine
src/utils/util_training.py - Learning metrics
src/metrics/metrics_learning.py - Metric logging
src/metrics/metrics_log.py - Model checkpointing
src/models/model_callbacks.py - PDF reports
src/metrics/report.py
Experiment management
- Hyperparameter search
src/utils/ultils_run.py - Batch experiment plan
src/batch_experiments/setup01.py - Batch runner
src/run_batch.py
Other components
- Researcher
- Weights & Biases
- FI-2010 files
Connections 33
- Researcher → CLI entry: starts run
README.md - CLI entry → Simulation: creates simulation
README.md - Simulation → Settings: configures
src/lobcast.py - Simulation → Hyperparameters: sets parameters
src/lobcast.py - Simulation → Dataset selection: selects dataset
src/lobcast.py - Dataset selection → Data module: uses module
src/data_preprocessing/utils_dataset.py - Dataset selection → FI-2010 samples: builds samples
src/data_preprocessing/utils_dataset.py - FI-2010 samples → FI-2010 files: reads files
src/data_preprocessing/FI/FIDataBuilder.py - Simulation → Model selection: selects model
src/lobcast.py - Model selection → Learning engine: wraps for training
src/models/utils_models.py - Model selection → Model registry: resolves model
src/models/utils_models.py - Model registry → MLP: registers
src/models/models_classes.py - Model registry → CNN1: registers
src/models/models_classes.py - Model registry → CNN2: registers
src/models/models_classes.py - Model registry → BINCTABL: registers
src/models/models_classes.py - MLP → Model interface: extends
src/models/mlp/mlp.py - CNN1 → Model interface: extends
src/models/cnn1/cnn1.py - CNN2 → Model interface: extends
src/models/cnn2/cnn2.py - BINCTABL → Model interface: extends
src/models/binctabl/binctabl.py - BINCTABL → BINCTABL layers: uses layers
src/models/binctabl/binctabl.py - Simulation → Learning engine: trains and tests
src/lobcast.py - Learning engine → Learning metrics: computes metrics
src/utils/util_training.py - Learning engine → Metric logging: records metrics
src/utils/util_training.py - Simulation → Model checkpointing: configures checkpointing
src/lobcast.py - Simulation → PDF reports: generates reports
src/lobcast.py - PDF reports → Shared utilities: reads JSON
src/metrics/report.py - Simulation → Hyperparameter search: uses sweep utilities
src/lobcast.py - CLI entry → Hyperparameter search: chooses configurations
README.md - Researcher → Batch runner: starts batch
README.md - Batch runner → Batch experiment plan: loads plan
src/run_batch.py - Batch runner → Simulation: runs configurations
README.md - Hyperparameter search → Weights & Biases: starts sweep
src/utils/ultils_run.py - Simulation → Weights & Biases: logs metrics
src/lobcast.py