How sebastien-doyez2812/openml-core works
OpenML-core is a PyTorch computer-vision modeling package. Its principal workflow is to create a model, train it on batches using a configured loss and optimizer, then predict, evaluate, and save or load weights. The sampled implementation supports a U-Net and reusable loss functions; README capabilities for other architectures and integrations are documented but not visible in the sampled source, so their wiring is not inferred.
Main entry points: base_model.py BaseModel · UNets.py U-Net network · loss.py LossFactory · convolutions.py DoubleConv blocks
Components 16
Model API
- BaseModel
src/openmlcore/models/base_model.py
Model Architecture
- UNet wrapper
src/openmlcore/models/UNets.py - U-Net network
src/openmlcore/models/UNets.py - DoubleConv blocks
src/openmlcore/models/lib/convolutions.py
Training Components
- LossFactory
src/openmlcore/loss/loss.py - Loss functions
src/openmlcore/loss/loss.py - Custom DataLoader
src/openmlcore/base/data.py
Documented Capabilities
- Other architectures
src/openmlcore/models - LoRA fine-tuning
src/openmlcore/models - Hugging Face Hub
- MLflow tracking
Other components
- Model user
- Training data
- Batch loader
- Configured optimizer
- Checkpoint files
Connections 17
- Model user → BaseModel: configures and invokes
- Model user → UNet wrapper: selects architecture
- UNet wrapper → U-Net network: creates
- U-Net network → DoubleConv blocks: uses blocks
- Model user → Batch loader: supplies batches
- Batch loader → BaseModel: yields batches
- BaseModel → U-Net network: runs forward
- BaseModel → Loss functions: computes training loss
- BaseModel → Configured optimizer: updates parameters
- Model user → LossFactory: selects loss
- LossFactory → Loss functions: creates
- BaseModel → Checkpoint files: saves and loads weights
- Model user → Custom DataLoader: may use
- Model user → Other architectures: selects documented models
- Model user → LoRA fine-tuning: uses documented fine-tuning
- Model user → Hugging Face Hub: pushes or pulls checkpoints
- Model user → MLflow tracking: tracks experiments