Very experimental add some way to get VNNLIB

DOCUMENTATION_SUMMARY.md

@@ -0,0 +1,151 @@
+# Documentation Updates Summary
+
+## Overview
+
+Updated documentation across Keras, PyTorch, and ONNX modules to guide users on the recommended workflow for neural networks with skip connections and residual architectures.
+
+## Key Message
+
+**For models with complex architectures (skip connections, ResNet, etc.):**
+- Export to ONNX format first
+- Use `add_onnx_dag_constr` for full DAG support
+- Direct Keras/PyTorch APIs only support sequential models
+
+## Files Updated
+
+### 1. Keras Documentation (`src/gurobi_ml/keras/keras.py`)
+
+**Added Section:** "Models with Skip Connections or Residual Architectures"
+
+**Content:**
+- Explains when to use ONNX export (skip connections, Functional API, etc.)
+- Two export methods with code examples:
+  - tf2onnx (recommended)
+  - keras2onnx (alternative)
+- Complete workflow from export to Gurobi ML usage
+- Links to external documentation
+
+**Key References:**
+- https://github.com/onnx/tensorflow-onnx (tf2onnx)
+- https://github.com/onnx/keras-onnx (keras2onnx)
+- https://onnx.ai/get-started.html (ONNX)
+
+### 2. PyTorch Documentation (`src/gurobi_ml/torch/sequential.py`)
+
+**Added Section:** "Models with Skip Connections or Residual Architectures"
+
+**Content:**
+- Explains limitations of torch.nn.Sequential
+- Complete torch.onnx.export example with all parameters
+- Model verification steps
+- Workflow from export to Gurobi ML usage
+- Links to official tutorials
+
+**Key References:**
+- https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html
+- https://pytorch.org/docs/stable/onnx.html
+- https://onnx.ai/get-started.html
+
+### 3. ONNX Module Documentation (`src/gurobi_ml/onnx/__init__.py`)
+
+**Enhanced:** Module-level docstring
+
+**Content:**
+- Clear distinction between add_onnx_constr and add_onnx_dag_constr
+- Quick reference examples for Keras and PyTorch export
+- Recommended workflow overview
+
+### 4. Comprehensive Guide (`ONNX_EXPORT_GUIDE.md`)
+
+**New File:** Complete tutorial and reference
+
+**Sections:**
+1. Overview - When to use ONNX export
+2. Keras Models with Skip Connections
+   - Full example with skip connection
+   - Step-by-step export process
+   - Gurobi ML integration
+3. PyTorch Models with Residual Connections
+   - ResNet-style example
+   - Detailed export parameters
+   - Accuracy verification
+4. Supported ONNX Operations
+5. Accuracy Information
+6. Installation Requirements
+7. Common Issues and Solutions
+8. References
+
+## Documentation Style
+
+All documentation follows these principles:
+
+1. **Problem First:** Explains when users need ONNX export
+2. **Step-by-Step:** Clear numbered steps with code examples
+3. **Complete Code:** Runnable examples, not fragments
+4. **External Links:** References to official documentation
+5. **Troubleshooting:** Common issues and solutions
+
+## Example Workflow Documented
+
+### Keras
+```python
+# 1. Export to ONNX
+import tf2onnx
+spec = (tf.TensorSpec((None, input_dim), tf.float32, name="input"),)
+model_proto, _ = tf2onnx.convert.from_keras(keras_model, input_signature=spec)
+onnx.save(model_proto, "model.onnx")
+
+# 2. Use with Gurobi ML
+from gurobi_ml.onnx import add_onnx_dag_constr
+onnx_model = onnx.load("model.onnx")
+pred = add_onnx_dag_constr(gp_model, onnx_model, input_vars)
+```
+
+### PyTorch
+```python
+# 1. Export to ONNX
+torch.onnx.export(pytorch_model, dummy_input, "model.onnx",
+                  export_params=True, opset_version=11)
+
+# 2. Use with Gurobi ML
+from gurobi_ml.onnx import add_onnx_dag_constr
+onnx_model = onnx.load("model.onnx")
+pred = add_onnx_dag_constr(gp_model, onnx_model, input_vars)
+```
+
+## Impact
+
+**Before:**
+- Users with ResNet/skip connection models would get NoModel errors
+- No guidance on alternatives
+- Frustration and confusion
+
+**After:**
+- Clear explanation of limitations
+- Step-by-step ONNX export workflow
+- Links to official documentation
+- Working examples
+- Troubleshooting guide
+
+## Testing
+
+All documentation has been verified for:
+- ✓ Python syntax correctness
+- ✓ Presence of key sections (Skip Connections, ONNX export)
+- ✓ References to export functions (tf2onnx, torch.onnx.export)
+- ✓ Links to external documentation
+- ✓ Complete code examples
+
+## Related Files
+
+- `DAG_IMPLEMENTATION_SUMMARY.md` - Technical implementation details
+- `COMPLETE_SOLUTION_SUMMARY.md` - Complete solution overview
+- `DEMO_DAG_IMPLEMENTATION.py` - Working demonstration script
+
+## Future Considerations
+
+If direct Keras/PyTorch DAG support is added later:
+1. Update these documentation sections
+2. Add new API functions
+3. Keep ONNX workflow as alternative
+4. Maintain backward compatibility