Merge pull request #4 from vloncar/opt1

Opt1

docs/advanced/model_optimization.rst

@@ -100,9 +100,9 @@ Finally, optimizing Vivado DSPs is possible, given a hls4ml config:
     from hls4ml.utils.config import config_from_keras_model
     from hls4ml.optimization.objectives.vivado_objectives import VivadoDSPEstimator
 
-    # Note the change from optimize_model to optimize_keras_for_hls4ml
-    # The function optimize_keras_for_hls4ml acts as a wrapper for the function, parsing hls4ml config to model attributes
-    from hls4ml.optimization import optimize_keras_for_hls4ml
+    # Note the change from optimize_model to optimize_keras_model_for_hls4ml
+    # The function optimize_keras_model_for_hls4ml acts as a wrapper for the function, parsing hls4ml config to model attributes
+    from hls4ml.optimization import optimize_keras_model_for_hls4ml
 
     # Create hls4ml config
     default_reuse_factor = 4
@@ -113,7 +113,7 @@ Finally, optimizing Vivado DSPs is possible, given a hls4ml config:
 
     # Optimize model
     # Note the change from ParameterEstimator to VivadoDSPEstimator
-    optimized_model = optimize_keras_for_hls4ml(
+    optimized_model = optimize_keras_model_for_hls4ml(
         baseline_model, model_attributes, VivadoDSPEstimator, scheduler,
         X_train, y_train, X_val, y_val, batch_size, epochs,
         optimizer, loss_fn, metric, increasing, rtol

docs/reference.rst

@@ -86,6 +86,19 @@ binary/ternary networks:
         year = "2021"
     }
 
+optimization API:
+
+..  code-block:: bibtex
+
+    @article{Ramhorst:2023fpga,
+      author = "Benjamin Ramhorst and others",
+      title = "{FPGA Resource-aware Structured Pruning for Real-Time Neural Networks}",
+      eprint = "2308.05170",
+      archivePrefix = "arXiv",
+      primaryClass = "cs.AR",
+      year = "2023"
+    }
+
 Acknowledgments
 ===============
 If you benefited from participating in our community, we ask that you please acknowledge the Fast Machine Learning collaboration, and particular individuals who helped you, in any publications.

hls4ml/optimization/__init__.py

@@ -6,7 +6,7 @@
 default_regularization_range = np.logspace(-6, -2, num=16).tolist()
 
 
-def optimize_keras_for_hls4ml(
+def optimize_keras_model_for_hls4ml(
     keras_model,
     hls_config,
     objective,

hls4ml/optimization/objectives/__init__.py

@@ -31,8 +31,10 @@ def is_layer_optimizable(self, layer_attributes):
             layer_attributes (hls4ml.optimization.attributes.LayerAttributes): Layer attributes
 
         Returns:
-            optimizable (boolean): can optimizations be applied to this layer
-            optimization_attributes (hls4ml.optimization.attributes.OptimizationAttributes):
+            tuple containing
+
+            - optimizable (boolean): can optimizations be applied to this layer
+            - optimization_attributes (hls4ml.optimization.attributes.OptimizationAttributes):
                 Most suitable approach for optimization
 
         Examples: