Merge pull request #5 from tilleyd/feature/nd-optimization

Optimize all functions for 2D arrays

LICENSE.txt

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2020 Duncan Tilley
+Copyright (c) 2022 Duncan Tilley
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,5 +1,7 @@
 # CEC 2017 Python
 
+> **⚠ 23 Nov 2022**: Breaking changes were made to all function signatures. See the changelog for details.
+
 Python 3 module containing a native implementation of the CEC 2017 benchmark functions (single objective optimization). The implementation is adapted from Awad's original C implementation, available on [Suganthan's GitHub repo](https://github.com/P-N-Suganthan/CEC2017-BoundContrained), along with the problem definitions [1].
 
 Although there are wrappers for the C code, this module is easier to use, natively supports numpy arrays and is (_much_) more readable than the C implementation.
@@ -14,9 +16,9 @@ As per the problem definitions, functions are defined for 10, 30, 50 and 100 dim
 
 ## Features
 
-- Native implementation of all CEC 2017 single objective functions
-- Pre-defined rotations, shifts and shuffles for 2, 10, 20, 30, 50 and 100 dimensions
-- Allows custom rotations, shifts and shuffles
+- Native implementation of all CEC 2017 single objective functions optimized for multiple simultaneous evaluations
+- Pre-defined rotations, shifts, and shuffles for 2, 10, 20, 30, 50 and 100 dimensions
+- Allows custom rotations, shifts, and shuffles
 - Convenient surface plot utility
 - Easy access to basic functions f1 to f19 (e.g. Ackley, Discus, etc.)
 
@@ -30,26 +32,38 @@ python3 setup.py install
 
 ## Usage
 
-Below is a simple example of executing either a single function or all functions. See [example.py](example.py) for more advanced use cases.
+Below is a simple example of executing either a single function or all functions. Note that each function takes a 2D array and returns a 1D array. See [example.py](example.py) for more advanced use cases.
 
 ```py
 # Using only f5:
 from cec2017.functions import f5
-x = np.random.uniform(-100, 100, size=50)
+samples = 3
+dimension = 50
+x = np.random.uniform(-100, 100, size=(samples, dimension))
 val = f5(x)
-print('f5(x) = %.6f' %val)
+for i in range(samples):
+    print(f"f5(x_{i}) = {val[i]:.6f}")
 
 # Using all functions:
 from cec2017.functions import all_functions
 for f in all_functions:
-    x = np.random.uniform(-100, 100, size=50)
+    x = np.random.uniform(-100, 100, size=(samples, dimension))
     val = f(x)
-    print('%s(x) = %.6f' %( f.__name__, val ))
+    for i in range(samples):
+        print(f"{f.__name__}(x_{i}) = {val[i]:.6f}")
 ```
 
+## Changelog
+
+### 23 Nov 2022
+- All functions have been reimplemented using numpy vector operations instead of native python loops.
+- **Breaking change:** Functions now expect a 2D input array of shape (m, D) where D is the dimensionality, and m is an arbitrary sample size.
+
+The above updates have allowed a substantial increase in performance when evaluating multiple parameter samples simultaneously, which previously would have had to be done as consecutive calls. See [PR 5](https://github.com/tilleyd/cec2017-py/pull/5) for more details.
+
 ## License
 
-Copyright © 2020 Duncan Tilley
+Copyright © 2022 Duncan Tilley
 See the [license notice](LICENSE.txt) for full details.
 
 ## Issues