Add graph-based surrogates for architecture optimization

docs/index.md

@@ -229,6 +229,16 @@ if __name__ == '__main__':
     DemoArchOptProblem().print_stats()
 ```
 
+A previously generated reference front can be stored as a NumPy array and loaded for use by experiment metrics:
+
+```python
+from sb_arch_opt.pareto_front import load_pareto_front
+
+reference_pf = load_pareto_front('reference_front/pf.npy')
+# Equivalently, pass the directory containing pf.npy:
+reference_pf = load_pareto_front('reference_front')
+```
+
 ### Architecture Optimization Problem with an Explicit Design Space Model
 
 If you are implementing an architecture optimization problem with design space hierarchy (i.e. some of your design

mkdocs.yml

@@ -28,6 +28,7 @@ nav:
     - 'SMARTy': 'algo/smarty.md'
   - Tutorials:
       - SBArchOpt: tutorial.ipynb
+      - Graph-based Optimization: guide_graph_based_optimization.ipynb
   - 'Test Problems': 'test_problems.md'
   - API Reference:
     - Problem Definition: 'api/problem.md'

sb_arch_opt/algo/arch_sbo/__init__.py

@@ -1 +1,2 @@
 from .api import *
+from .graph import *

sb_arch_opt/algo/arch_sbo/graph/__init__.py

@@ -0,0 +1,7 @@
+from .representation import *
+from .kernels import *
+
+from sb_arch_opt.algo.arch_sbo.models import HAS_SMT
+
+if HAS_SMT:
+    from .surrogate import *