Replace LH by CLH as optimization algorithm.

Project.toml

@@ -1,7 +1,7 @@
 name = "PotentialLearning"
 uuid = "82b0a93c-c2e3-44bc-a418-f0f89b0ae5c2"
 authors = ["CESMIX Team"]
-version = "0.2.3"
+version = "0.2.4"
 
 [deps]
 AtomsBase = "a963bdd2-2df7-4f54-a1ee-49d51e6be12a"

examples/Opt-ACE-aHfO2/fit-opt-ace-ahfo2.jl

@@ -52,12 +52,12 @@ end;
 model = ACE
 pars = OrderedDict( :body_order        => [2, 3, 4],
                     :polynomial_degree => [3, 4, 5],
-                    :rcutoff           => [4.5, 5.0, 5.5],
-                    :wL                => [0.5, 1.0, 1.5],
-                    :csp               => [0.5, 1.0, 1.5],
-                    :r0                => [0.5, 1.0, 1.5]);
+                    :rcutoff           => LinRange(3.5, 6.5, 10),
+                    :wL                => LinRange(0.3, 1.8, 10),
+                    :csp               => LinRange(0.3, 1.8, 10),
+                    :r0                => LinRange(0.3, 1.8, 10));
 
-# Use random sampling to find the optimal hyper-parameters.
+# Use **random sampling** to find the optimal hyper-parameters.
 iap, res = hyperlearn!(model, pars, conf_train; 
                        n_samples = 10, sampler = RandomSampler(),
                        loss = custom_loss, ws = [1.0, 1.0], int = true);
@@ -74,21 +74,21 @@ err_time = plot_err_time(res)
 @save_fig res_path err_time
 DisplayAs.PNG(err_time)
 
-
-# Alternatively, use latin hypercube sampling to find the optimal hyper-parameters.
-iap, res = hyperlearn!(model, pars, conf_train; 
-                       n_samples = 3, sampler = LHSampler(),
-                       loss = custom_loss, ws = [1.0, 1.0], int = true);
+# Alternatively, use **latin hypercube sampling** to find the optimal hyper-parameters.
+sampler = CLHSampler(dims=[Categorical(3), Categorical(3), Continuous(),
+                           Continuous(), Continuous(), Continuous()])
+iap2, res2 = hyperlearn!(model, pars, conf_train; 
+                         n_samples = 10, sampler = sampler,
+                         loss = custom_loss, ws = [1.0, 1.0], int = true);
 
 # Save and show results.
-@save_var res_path iap.β
-@save_var res_path iap.β0
-@save_var res_path iap.basis
-@save_dataframe res_path res
-res
+@save_var res_path iap2.β
+@save_var res_path iap2.β0
+@save_var res_path iap2.basis
+@save_dataframe res_path res2
+res2
 
 # Plot error vs time.
-err_time = plot_err_time(res)
-@save_fig res_path err_time
-DisplayAs.PNG(err_time)
-
+err_time2 = plot_err_time(res2)
+@save_fig res_path err_time2
+DisplayAs.PNG(err_time2)

src/HyperLearning/linear-hyperlearn.jl

@@ -57,9 +57,9 @@ function hyperlearn!(
     model::DataType,
     pars::OrderedDict,
     conf_train::DataSet;
-    n_samples = 5,
+    n_samples = 10,
     sampler = RandomSampler(),
-    loss = loss,
+    loss = hyperlearn!,
     ws = [1.0, 1.0],
     int = true
 )