Update Installation Instructions and Python 3 Compatibility

README.md

@@ -1,25 +1,26 @@
-Summary
-=======
+# Summary
+
 This repository contains some basic code for Bayesian optimization
 
-Installation
-============
 
-Install the current development version of scikit-learn (or sklearn version 0.18 once this is available)
+## Installation
 
-    git clone git@github.com:scikit-learn/scikit-learn.git
-    cd sklearn
-    sudo python setup.py install
+Install dependencies
+
+1. Install [nlopt](https://github.com/stevengj/nlopt) for your Python version.
+2. Install dependencies with pip: `[sudo] pip[3] install -r requirements.txt`
+   (requirements.txt can be found in the repository)
+3. (Optional:) install [BOLeRo](https://github.com/rock-learning/bolero)
 
 Install `bayesian_optimization`
 
-    git clone git@git.hb.dfki.de:jmetzen/bayesian_optimization.git
+    git clone https://github.com/rock-learning/bayesian_optimization.git
     cd bayesian_optimization
     sudo python setup.py install
 
 
-Usage
-=====
+## Usage
+
 Some usage examples are contained in the folder "examples". To reproduce the results from the ICML 2016 paper
 "Minimum Regret Search for Single- and Multi-Task Optimization", please execute the jupyter notebook "examples/mrs_evaluation.ipynb." 
 

bayesian_optimization/__init__.py

@@ -1,6 +1,7 @@
-from bayesian_optimization import (BayesianOptimizer, REMBOOptimizer,
-    InterleavedREMBOOptimizer)
-from model import GaussianProcessModel
-from acquisition_functions import (ProbabilityOfImprovement,
-    ExpectedImprovement, UpperConfidenceBound, GPUpperConfidenceBound,
-    EntropySearch, MinimalRegretSearch, create_acquisition_function)
+from .bayesian_optimization import (BayesianOptimizer, REMBOOptimizer,
+                                    InterleavedREMBOOptimizer)
+from .model import GaussianProcessModel
+from .acquisition_functions import (
+    ProbabilityOfImprovement, ExpectedImprovement, UpperConfidenceBound,
+    GPUpperConfidenceBound, EntropySearch, MinimalRegretSearch,
+    create_acquisition_function)

bayesian_optimization/acquisition_functions.py

@@ -5,9 +5,6 @@
 from scipy.stats import norm, entropy
 from scipy.special import erf
 
-from sklearn.cluster import KMeans
-from sklearn.neighbors import NearestNeighbors
-
 
 class AcquisitionFunction(object):
     """ Abstract base class for acquisition functions."""

bayesian_optimization/bayesian_optimization.py

@@ -333,7 +333,7 @@ def __init__(self, interleaved_runs=2, *args, **kwargs):
                                       *args, **kwargs)
                        for run in range(interleaved_runs)]
         self.rembos = cycle(self.rembos)
-        self.current_rembo = self.rembos.next()
+        self.current_rembo = next(self.rembos)
 
         self.X_ = []
         self.y_ = []
@@ -361,4 +361,4 @@ def update(self, X, y):
         self.y_.append(y)
 
         self.current_rembo.update(X, y)
-        self.current_rembo = self.rembos.next()
+        self.current_rembo = next(self.rembos)

bolero_bayes_opt/examples/plot_bops_optimizer_catapult.py

@@ -17,13 +17,13 @@
 
 
 kernel = C(100.0, (1.0, 10000.0)) \
-    * Matern(l=(1.0, 1.0), l_bounds=[(0.1, 100), (0.1, 100)])
+    * Matern(length_scale=(1.0, 1.0), length_scale_bounds=[(0.1, 100), (0.1, 100)])
 
 opt = BOPSOptimizer(
     boundaries=[(5, 10), (0, np.pi/2)], bo_type="bo",
     acquisition_function="UCB", acq_fct_kwargs=dict(kappa=2.5),
     optimizer="direct+lbfgs", maxf=100,
-    gp_kwargs=dict(kernel=kernel, normalize_y=True, sigma_squared_n=1e-5))
+    gp_kwargs=dict(kernel=kernel, normalize_y=True, alpha=1e-5))
 
 
 target = 4.0  # Fixed target
@@ -42,7 +42,7 @@
 
 # Determine reward landscape
 v = np.linspace(5.0, 10.0, 100)
-theta = np.linspace(0.0, np.pi/2, 100)
+theta = np.linspace(0.0, np.pi / 2, 100)
 V, Theta = np.meshgrid(v, theta)
 
 Z = np.array([[partial(catapult._compute_reward, context=context)

bolero_bayes_opt/optimizer/bocps.py

@@ -84,7 +84,7 @@ def __init__(self, boundaries, bo_type="bo", acquisition_function="UCB",
         self.boundaries = boundaries
         self.bo_type = bo_type
         self.value_transform = value_transform
-        if isinstance(self.value_transform, basestring):
+        if isinstance(self.value_transform, str):
             self.value_transform = eval(self.value_transform)
         self.optimizer = optimizer
 
@@ -171,7 +171,7 @@ def set_evaluation_feedback(self, feedbacks):
             self.policy_fitted = False
 
     def best_policy(self, maxfun=15000, variance=0.01,
-                    training=["model-free", "model_based"]):
+                    training=["model-free", "model-based"]):
         """Returns the best (greedy) policy learned so far.
 
         Parameters

requirements.txt

@@ -0,0 +1,2 @@
+scikit-learn
+scipy