v0.2.0

.gitignore

@@ -8,7 +8,13 @@ __pycache__/
 # C extensions
 *.so
 
+# Local files
+IBMPlexMono-Regular.ttf
 .vscode
+**.DS_Store
+**paper
+*.png
+*.ipynb
 
 # Distribution / packaging
 .Python

bolift/__init__.py

@@ -1,6 +1,6 @@
 from .version import __version__
 from .asktell import AskTellFewShotMulti, AskTellFewShotTopk
-from .asktellfinetuning import AskTellFinetuning
+from .asktellFinetuning import AskTellFinetuning
 
 try:
     from .asktellGPR import AskTellGPR

bolift/aqfxns.py

@@ -10,6 +10,20 @@ def expected_improvement(dist, best):
     elif isinstance(dist, GaussDist):
         return expected_improvement_g(dist.mean(), dist.std(), best)
 
+def log_expected_improvement(dist, best):
+    """Log Expected improvement for the given discrete distribution"""
+    if isinstance(dist, DiscreteDist):
+        return log_expected_improvement_d(dist.probs, dist.values, best)
+    elif isinstance(dist, GaussDist):
+        return log_expected_improvement_g(dist.mean(), dist.std(), best)
+
+# I think it's just taking the log of the final EI computation. Will test this later
+# def log_expected_improvement(dist, best):
+#     """Log Expected improvement for the given discrete distribution"""
+#     if isinstance(dist, DiscreteDist):
+#         return np.log(expected_improvement_d(dist.probs, dist.values, best))
+#     elif isinstance(dist, GaussDist):
+#         return np.log(expected_improvement_g(dist.mean(), dist.std(), best))
 
 def probability_of_improvement(dist, best):
     """Probability of improvement for the given discrete distribution"""
@@ -40,6 +54,11 @@ def expected_improvement_d(probs, values, best):
     ei = np.sum(np.maximum(values - best, 0) * probs)
     return ei
 
+def log_expected_improvement_d(probs, values, best):
+    """Log Expected improvement for the given discrete distribution"""
+    # ei = np.sum(np.maximum(values - best, 0) * probs)
+    log_ei = np.log(np.sum(np.maximum(values - best, 0) * probs)+1e-15)
+    return log_ei
 
 def probability_of_improvement_d(probs, values, best):
     """Probability of improvement for the given discrete distribution"""
@@ -65,6 +84,13 @@ def expected_improvement_g(mean, std, best):
     ei = (mean - best) * norm.cdf(z) + std * norm.pdf(z)
     return ei
 
+def log_expected_improvement_g(mean, std, best):
+    """Log Expected improvement for the given Gaussian distribution"""
+    z = (mean - best) / std 
+    # ei = std * h(z)
+    # ei = std * (norm.pdf(z) + z * norm.cdf(z))
+    log_ei = np.log(std) + np.log((norm.pdf(z) + z * norm.cdf(z)))
+    return log_ei
 
 def probability_of_improvement_g(mean, std, best):
     """Probability of improvement for the given Gaussian distribution"""