From 925c2f9c208e144e6ca9f16e94a8db64ced2ea0d Mon Sep 17 00:00:00 2001
From: Israel Figueroa <israel.figueroa@usm.cl>
Date: Sun, 16 Mar 2025 22:24:57 -0300
Subject: [PATCH] tune

---
 TrainerClass.py | 27 +++++++++++++++++----------
 1 file changed, 17 insertions(+), 10 deletions(-)

diff --git a/TrainerClass.py b/TrainerClass.py
index a5b310d..2d0e8e7 100644
--- a/TrainerClass.py
+++ b/TrainerClass.py
@@ -184,12 +184,16 @@ class eNoseTrainer:
         X_test_ref = ray.put(X_test_orig)
         Y_test_ref = ray.put(Y_test_orig)
 
-
+# input channels 80 window
+# l1 minimo 2, maximo 5 => 40 | 16
+# maxpool 2 o 3 => 20 |
+# l2 minimo 2 maximo 5 => 4 | 2/5
+# maxpool 2 o 3 => 2 | 2/15
         def build_model_conv1D(config, input_shape, output_dim):
             model = keras.Sequential([
-                layers.Conv1D(filters=config['filters'], kernel_size=config['kernel_size'], stride=config['kernel_size']//2, activation='relu', input_shape=input_shape),
+                layers.Conv1D(filters=config['filters'], kernel_size=config['kernel_l1'], stride=config['kernel_l1']//2, activation='relu', padding='causal', input_shape=input_shape),
                 layers.MaxPooling1D(pool_size=config['pool_size']),
-                layers.Conv1D(filters=config['filters'] * 2, kernel_size=config['kernel_size'], stride=config['kernel_size']//2, activation='relu'),
+                layers.Conv1D(filters=config['filters'] * 2, kernel_size=config['kernel_l2'], stride=config['kernel_l2']//2, activation='relu', padding='causal'),
                 layers.MaxPooling1D(pool_size=config['pool_size']),
                 layers.Flatten(),
                 layers.Dense(config['dense_units'], activation='relu'),
@@ -226,15 +230,18 @@ class eNoseTrainer:
 
         config_space = {
             'filters': tune.choice([16, 32, 64]),
-            'kernel_size': tune.choice([3, 5, 7]),
+            'kernel_l1': tune.choice([3, 5, 7]),
+            'kernel_l2': tune.choice([3, 5, 7]),
             'pool_size': tune.choice([2, 3]),
-            'dense_units': tune.choice([32, 64, 128]),
-            'dropout': tune.choice([0.1, 0.2, 0.3]),
-            'lr': tune.choice([0.001, 0.0005, 0.0001]),
+            'dense_units': tune.choice([32, 64, 128, 256]),
+            'dropout': tune.choice([0.05, 0.15, 0.3]),
+            'lr': tune.choice([0.01, 0.005, 0.001]),
             'batch_size': tune.choice([16, 32, 64]),
-            'epochs': epochs
+            'epochs': epochssample_space
         }
-        total_space = (3*3*2*3*3*3*3)
+        total_space = (3*3*3*2*3*3*3*3)
+
+        self.logger.debug(f"total_space: {total_space}, num_samples: {int(nsamples*total_space)}")
 
         scheduler = ASHAScheduler(metric='mse', mode='min', max_t=epochs, grace_period=5, reduction_factor=2)
         # analysis = tune.run(train_model, config=config_space, num_samples=num_samples, scheduler=scheduler)
@@ -354,7 +361,7 @@ class eNoseTrainer:
             plt.close()
 
     def fit(self):
-        windows = [32, 64, 128]
+        windows = [128, 256, 384]
         total_train_queue = 2*int(1/self.ratio)*(len(self.get_model_train())+len(windows))
         self.logger.info("{:=^60}".format(f'Begin Fit {total_train_queue} Models'))
         self.trained = 0