Acceleration

Author: yuta-nakahara

bayesml/bernoulli/_bernoulli.py

@@ -386,6 +386,12 @@ def calc_pred_dist(self):
         """Calculate the parameters of the predictive distribution."""
         self.p_theta = self.hn_alpha / (self.hn_alpha + self.hn_beta)
 
+    def _calc_pred_density(self,x):
+        if x:
+            return self.p_theta
+        else:
+            return 1.0-self.p_theta
+
     def make_prediction(self,loss="squared"):
         """Predict a new data point under the given criterion.
 

bayesml/exponential/_exponential.py

@@ -382,6 +382,9 @@ def calc_pred_dist(self):
         self.p_kappa = self.hn_alpha
         self.p_lambda = self.hn_beta
 
+    def _calc_pred_density(self,x):
+        return ss_lomax.pdf(x,c=self.p_kappa,scale=self.p_lambda)
+
     def make_prediction(self,loss="squared"):
         """Predict a new data point under the given criterion.
 

bayesml/metatree/_metatree.py

@@ -413,7 +413,7 @@ def _gen_sample_recursion(self,node:_Node,x_continuous,x_categorical):
         else:
             if node.k < self.c_dim_continuous:
                 for i in range(self.c_num_children_vec[node.k]):
-                    if node.thresholds[i] < x_continuous[node.k] and x_continuous[node.k] < node.thresholds[i+1]:
+                    if x_continuous[node.k] < node.thresholds[i+1]:
                         index = i
                         break
             else:
@@ -1698,40 +1698,26 @@ def _copy_tree_from_sklearn_tree(self,new_node:_Node, original_tree,node_id):
             new_node.h_g = 0.0
             new_node.leaf = True
 
-    def _update_posterior_leaf(self,node:_Node,x_continuous,y):
-            try:
-                node.sub_model.calc_pred_dist(x_continuous)
-            except:
-                node.sub_model.calc_pred_dist()
-            pred_dist = node.sub_model.make_prediction(loss='KL') # Futurework: direct method to get marginal likelihood is better
-
-            try:
-                node.sub_model.update_posterior(x_continuous,y)
-            except:
-                node.sub_model.update_posterior(y)
-
-            if type(pred_dist) is np.ndarray:
-                return pred_dist[y]
-            try:
-                return pred_dist.pdf(y)
-            except:
-                return pred_dist.pmf(y)
+    def _update_posterior_leaf(self,node:_Node,y):
+        node.sub_model.calc_pred_dist()
+        node.sub_model.update_posterior(y)
+        return node.sub_model._calc_pred_density(y)
 
     def _update_posterior_recursion(self,node:_Node,x_continuous,x_categorical,y):
         if not node.leaf:  # inner node
             if node.k < self.c_dim_continuous:
                 for i in range(self.c_num_children_vec[node.k]):
-                    if node.thresholds[i] < x_continuous[node.k] and x_continuous[node.k] < node.thresholds[i+1]:
+                    if x_continuous[node.k] < node.thresholds[i+1]:
                         index = i
                         break
             else:
                 index = x_categorical[node.k-self.c_dim_continuous]
             tmp1 = self._update_posterior_recursion(node.children[index],x_continuous,x_categorical,y)
-            tmp2 = (1 - node.h_g) * self._update_posterior_leaf(node,x_continuous,y) + node.h_g * tmp1
+            tmp2 = (1 - node.h_g) * self._update_posterior_leaf(node,y) + node.h_g * tmp1
             node.h_g = node.h_g * tmp1 / tmp2
             return tmp2
         else:  # leaf node
-            return self._update_posterior_leaf(node,x_continuous,y)
+            return self._update_posterior_leaf(node,y)
 
     def _compare_metatree_recursion(self,node1:_Node,node2:_Node):
         if node1.leaf:
@@ -2302,7 +2288,7 @@ def _calc_pred_dist_recursion(self,node:_Node,x_continuous,x_categorical):
         if not node.leaf:  # inner node
             if node.k < self.c_dim_continuous:
                 for i in range(self.c_num_children_vec[node.k]):
-                    if node.thresholds[i] < x_continuous[node.k] and x_continuous[node.k] < node.thresholds[i+1]:
+                    if x_continuous[node.k] < node.thresholds[i+1]:
                         index = i
                         break
             else:
@@ -2377,7 +2363,7 @@ def _make_prediction_recursion_squared(self,node:_Node):
         else:  # inner node
             if node.k < self.c_dim_continuous:
                 for i in range(self.c_num_children_vec[node.k]):
-                    if node.thresholds[i] < self._tmp_x_continuous[node.k] and self._tmp_x_continuous[node.k] < node.thresholds[i+1]:
+                    if self._tmp_x_continuous[node.k] < node.thresholds[i+1]:
                         index = i
                         break
             else:
@@ -2391,7 +2377,7 @@ def _make_prediction_recursion_kl(self,node:_Node):
         else:  # inner node
             if node.k < self.c_dim_continuous:
                 for i in range(self.c_num_children_vec[node.k]):
-                    if node.thresholds[i] < self._tmp_x_continuous[node.k] and self._tmp_x_continuous[node.k] < node.thresholds[i+1]:
+                    if self._tmp_x_continuous[node.k] < node.thresholds[i+1]:
                         index = i
                         break
             else:

bayesml/metatree/metatree_test.py

@@ -2,28 +2,35 @@
 from bayesml import normal
 from bayesml import poisson
 from bayesml import bernoulli
+from bayesml import exponential
 import numpy as np
-import copy
+import time
 
-dim_continuous = 0
-dim_categorical = 2
+dim_continuous = 2
+dim_categorical = 0
 
 gen_model = metatree.GenModel(
     c_dim_continuous=dim_continuous,
     c_dim_categorical=dim_categorical,
-    h_g=0.75,
-    sub_h_params={'h_alpha':0.1,'h_beta':0.1})
+    h_g=1.0,
+    SubModel=normal,
+    sub_h_params={'h_kappa':0.1})
+    # sub_h_params={'h_alpha':0.1,'h_beta':0.1})
 gen_model.gen_params(threshold_type='even')
-gen_model.visualize_model(filename='tree.pdf')
+# gen_model.visualize_model(filename='tree.pdf')
 
 x_continuous,x_categorical,y = gen_model.gen_sample(200)
-x_continuous_test,x_categorical_test,y_test = gen_model.gen_sample(10)
 
 learn_model = metatree.LearnModel(
     c_dim_continuous=dim_continuous,
     c_dim_categorical=dim_categorical,
     c_num_children_vec=2,
-    sub_h0_params={'h0_alpha':0.1,'h0_beta':0.1})
+    SubModel=normal,
+    sub_h0_params={'h0_kappa':0.1})
+    # sub_h0_params={'h0_alpha':0.1,'h0_beta':0.1})
+
+start = time.time()
 learn_model.update_posterior(x_continuous,x_categorical,y)
-for i in range(10):
-    print(learn_model.pred_and_update(x_continuous_test[i],x_categorical_test[i],y_test[i],loss='0-1'))
+end = time.time()
+
+print(end-start)

bayesml/normal/_normal.py

@@ -466,6 +466,9 @@ def calc_pred_dist(self):
         self.p_nu = 2*self.hn_alpha
         self.p_lambda = self.hn_kappa / (self.hn_kappa+1) * self.hn_alpha / self.hn_beta
 
+    def _calc_pred_density(self,x):
+        return ss_t.pdf(x,loc=self.p_mu,scale=1.0/np.sqrt(self.p_lambda),df=self.p_nu)
+
     def make_prediction(self,loss="squared"):
         """Predict a new data point under the given criterion.
 

bayesml/poisson/_poisson.py

@@ -377,6 +377,9 @@ def calc_pred_dist(self):
         self.p_r = self.hn_alpha
         self.p_theta = 1.0 / (1.0+self.hn_beta)
 
+    def _calc_pred_density(self,x):
+        return ss_nbinom.pmf(x,n=self.p_r,p=(1.0-self.p_theta))
+
     def make_prediction(self,loss="squared"):
         """Predict a new data point under the given criterion.