ltfmselector 0.1.11__py3-none-any.whl → 0.1.12__py3-none-any.whl

ltfmselector/ltfmselector.py

@@ -1,6 +1,7 @@
 import torch
 import torch.nn as nn
 import torch.optim as optim
+from torch.utils.tensorboard import SummaryWriter
 
 import os
 import random
@@ -226,7 +227,7 @@ class LTFMSelector:
     def fit(
             self, X, y, loss_function='mse', sample_weight=None,
             agent_neuralnetwork=None, lr=1e-5, returnQ=False,
-            background_dataset=None, **kwargs
+            monitor=False, background_dataset=None, **kwargs
     ):
         '''
         Initializes the environment and agent, then trains the agent to select
@@ -265,7 +266,14 @@ class LTFMSelector:
 
         returnQ : bool
             Return average computed action-value functions and rewards of
-            the sampled batches, for debugging purposes.
+            the sampled batches, as a (<total_iterations>, 3) matrix. The 
+            columns correspond to the averaged Q, reward, and target functions.
+
+        monitor : bool
+            Monitor training process using a TensorBoard.
+
+            Run `tensorboard --logdir=runs` in the terminal to monitor the p
+            progression of the action-value function.
 
         background_dataset : None or pd.DataFrame
             If None, numerical features will be assumed when computing the
@@ -312,10 +320,18 @@ class LTFMSelector:
         self.sample_weight = sample_weight
 
         # If user wants to monitor progression of terms in the loss function
+        if monitor:
+            writer = SummaryWriter()
+            monitor_count = 1
+
+        # If user wants to save average computed action-value functions and
+        # rewards of sampled batches
         if returnQ:
-            Q_avr_list = []
-            r_avr_list = []
-            V_avr_list = []
+            total_iterations = 10000000000
+            LearningValuesMatrix = np.zeros(
+                (total_iterations, 3), dtype=np.float32
+            )
+            Q_count = 1
 
         # Initializing the environment
         env = Environment(
@@ -406,13 +422,21 @@ class LTFMSelector:
                 state = next_state
 
                 # Optimize the model
-                _res = self.optimize_model(optimizer, loss_function, returnQ)
+                _res = self.optimize_model(optimizer, loss_function, monitor, returnQ)
+
+                if monitor:
+                    if not _res is None:
+                        writer.add_scalar("Metrics/Average_QValue", _res[0], monitor_count)
+                        writer.add_scalar("Metrics/Average_Reward", _res[1], monitor_count)
+                        writer.add_scalar("Metrics/Average_Target", _res[2], monitor_count)
+                        monitor_count += 1
 
                 if returnQ:
                     if not _res is None:
-                        Q_avr_list.append(_res[0])
-                        r_avr_list.append(_res[1])
-                        V_avr_list.append(_res[2])
+                        LearningValuesMatrix[Q_count, 0] = _res[0]
+                        LearningValuesMatrix[Q_count, 1] = _res[1]
+                        LearningValuesMatrix[Q_count, 2] = _res[2]
+                        Q_count += 1
 
                 # Apply soft update to target network's weights
                 targetParameters = self.target_net.state_dict()
@@ -456,11 +480,19 @@ class LTFMSelector:
                 self.policy_network_checkpoints[self.episodes] =\
                     self.policy_net.state_dict()
 
+        if monitor:
+            writer.add_scalar("Metrics/Average_QValue", _res[0], monitor_count)
+            writer.add_scalar("Metrics/Average_Reward", _res[1], monitor_count)
+            writer.add_scalar("Metrics/Average_Target", _res[2], monitor_count)
+            writer.close()
+
         if returnQ:
-            Q_avr_list.append(_res[0])
-            r_avr_list.append(_res[1])
-            V_avr_list.append(_res[2])
-            return doc, (Q_avr_list, r_avr_list, V_avr_list)
+            LearningValuesMatrix[Q_count, 0] = _res[0]
+            LearningValuesMatrix[Q_count, 1] = _res[1]
+            LearningValuesMatrix[Q_count, 2] = _res[2]
+
+        if (monitor or returnQ):
+            return doc, LearningValuesMatrix[0:Q_count+1, :]
         else:
             return doc
 
@@ -572,7 +604,7 @@ class LTFMSelector:
             with torch.no_grad():
                 return (self.policy_net(state).max(1)[1].view(1, 1) - 1)
 
-    def optimize_model(self, optimizer, loss_function, returnQ):
+    def optimize_model(self, optimizer, loss_function, monitor, returnQ):
         '''
         Optimize the policy network.
 
@@ -698,7 +730,7 @@ class LTFMSelector:
         # Optimize the model (policy network)
         optimizer.step()
 
-        if returnQ:
+        if (monitor or returnQ):
             Q_avr = state_action_values.detach().numpy().mean()
             r_avr = reward_batch.unsqueeze(1).numpy().mean()
             V_avr = expected_state_action_values.unsqueeze(1).numpy().mean()

{ltfmselector-0.1.11.dist-info → ltfmselector-0.1.12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ltfmselector
-Version: 0.1.11
+Version: 0.1.12
 Summary: Locally-Tailored Feature and Model Selector with Deep Q-Learning
 Project-URL: GitHub, https://github.com/RenZhen95/ltfmselector/
 Author-email: RenZhen95 <j-liaw@hotmail.com>
@@ -32,8 +32,9 @@ Requires-Dist: matplotlib>=3.10.1
 Requires-Dist: numpy>=2.2.4
 Requires-Dist: openpyxl>=3.1.5
 Requires-Dist: pandas>=2.2.3
-Requires-Dist: scikit-learn>=1.6.1
+Requires-Dist: scikit-learn<1.6
 Requires-Dist: seaborn>=0.13.2
+Requires-Dist: tensorboard>=2.20.0
 Requires-Dist: torch>=2.6.0
 Description-Content-Type: text/markdown
 

{ltfmselector-0.1.11.dist-info → ltfmselector-0.1.12.dist-info}/RECORD

@@ -1,9 +1,9 @@
 ltfmselector/__init__.py,sha256=lf3e90CNpEDvEmNZ-0iuoHOPsA7D-WN_opbBsTYLVEA,76
 ltfmselector/env.py,sha256=vizWGqDSc_2Zfs9aXjFARanIAz6PTKwUHu2_Lew9s3Y,13878
-ltfmselector/ltfmselector.py,sha256=-uRYcj89l8GL5re5Zw0mxe9tp8ulSmSBawqLA96S5A8,27984
+ltfmselector/ltfmselector.py,sha256=vs9unOmoDKq1piV6t87GC1wdy7kP8ucKHihw6i0F4KI,29567
 ltfmselector/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ltfmselector/utils.py,sha256=VXYZSDm7x4s0p9F_58NLW8WQa3dxi0vHZewRy6miC2E,5438
-ltfmselector-0.1.11.dist-info/METADATA,sha256=rPl6VeICmX_bw4B5QEZLlsXJcMVJ2-xGfJ9cZwTe3oA,2989
-ltfmselector-0.1.11.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-ltfmselector-0.1.11.dist-info/licenses/LICENSE,sha256=tmIDlkkp4a0EudXuGmeTdGjHjPhmmXkEMshACXLqX2w,1092
-ltfmselector-0.1.11.dist-info/RECORD,,
+ltfmselector-0.1.12.dist-info/METADATA,sha256=QaUPeSx9NlZx0ZUbkEPRyFS-8nfJz9Y8yV5TXXPc7fA,3021
+ltfmselector-0.1.12.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+ltfmselector-0.1.12.dist-info/licenses/LICENSE,sha256=tmIDlkkp4a0EudXuGmeTdGjHjPhmmXkEMshACXLqX2w,1092
+ltfmselector-0.1.12.dist-info/RECORD,,