Talos Classification Test

jacobbieker · Feb 9, 2019 · 4699ca8 · 4699ca8
1 parent 7f1c32e
commit 4699ca8
Show file tree

Hide file tree

Showing 2 changed files with 39 additions and 29 deletions.
diff --git a/examples/talos_classification.py b/examples/talos_classification.py
@@ -21,35 +21,37 @@
 # Parameter Dictionary for talos
 
 params = {'lr': (1, 10, 50),
-          'first_neuron': [4, 8, 16, 32, 64, 128, 256],
-          'last_neuron': [4, 8, 16, 32, 64, 128, 256],
-          'hidden_layers': [2, 3, 4, 5, 6, 7, 8],
-          'batch_size': [2, 3, 4, 8, 16, 32],
+          'first_neuron': [4, 16, 32],
+          'last_neuron': [4, 8, 16],
+          'hidden_layers': [2, 3, 4],
+          'batch_size': [2, 8, 32],
           'epochs': [500],
-          'dropout': (0, 0.80, 10),
+          'dropout': (0, 0.40, 5),
           'weight_regulizer': [None],
           'emb_output_dims': [None],
-          'optimizer': [adam, nadam, rmsprop, sgd],
+          'optimizer': [adam, nadam, rmsprop],
           'losses': [categorical_crossentropy, logcosh],
           'activation': [relu, elu],
           'last_activation': [softmax],
 
-          'neuron_1': [4, 8, 16, 32, 64],
-          'kernel_1': [1, 2, 3, 4, 5],
+          'neuron_1': [8, 16, 32, 64],
+          'kernel_1': [1, 3, 5],
           'stride_1': [1, 2, 3],
-          'rec_dropout': [0.0, 0.8, 40],
-          'rec_act': [relu, elu, hard_sigmoid, tanh, sigmoid],
-          'layer_drop': [0.0, 0.8, 40],
-          'pool': [0, 1],
-          'neuron_2': [4, 8, 16, 32, 64],
-          'kernel_2': [1, 2, 3],
-          'stride_2': [1, 2],
-          'three': [0, 1],
-          'neuron_3': [4, 8, 16, 32, 64],
-          'kernel_3': [1, 2, 3],
-          'stride_3': [1, 2],
-          }
+          'rec_dropout': [0.0, 0.4, 5],
+          'rec_act': [hard_sigmoid, tanh],
+          'layer_drop': [0.0, 0.4, 5],
 
+          }
+'''
+'pool': [0, 1],
+'neuron_2': [4, 8, 16, 32, 64],
+'kernel_2': [1, 2, 3],
+'stride_2': [1, 2],
+'three': [0, 1],
+'neuron_3': [4, 8, 16, 32, 64],
+'kernel_3': [1, 2, 3],
+'stride_3': [1, 2],
+'''
 
 def input_model(x_train, y_train, x_val, y_val, params):
     model = Sequential()
@@ -62,6 +64,7 @@ def input_model(x_train, y_train, x_val, y_val, params):
                          recurrent_activation=params['rec_act'],
                          return_sequences=False,
                          stateful=False))
+    '''
     if params['pool']:
         model.add(MaxPooling2D())
     model.add(Conv2D(params['neuron_2'], kernel_size=params['kernel_2'], strides=params['stride_2'],
@@ -73,6 +76,7 @@ def input_model(x_train, y_train, x_val, y_val, params):
                          padding='same', activation=params['activation']))
         model.add(MaxPooling2D())
         model.add(Dropout(model.add('layer_drop')))
+    '''
     model.add(Flatten())
 
     hidden_layers(model, params, params['last_neuron'])
@@ -87,23 +91,25 @@ def input_model(x_train, y_train, x_val, y_val, params):
                     batch_size=params['batch_size'],
                     epochs=params['epochs'], verbose=0,
                     validation_data=[x_val, y_val],
-                    callbacks=early_stopper(params['epochs'],
-                                            mode='moderate'))
+                    callbacks=[early_stopper(params['epochs'],
+                                            mode='moderate')])
 
     return out, model
 
 
-directory = "/home/jacob/Documents/iact_events/"
-gamma_dir = [directory + "gammaFeature/no_clean/"]
-proton_dir = [directory + "protonFeature/no_clean/"]
+directory = "/media/jacob/WDRed8TB1/Insync/iact_events/"
+gamma_dir = [directory + "gamma/no_clean/"]
+proton_dir = [directory + "proton/no_clean/"]
 
 x, y = get_chunk_of_data(directory=gamma_dir, proton_directory=proton_dir, indicies=(30, 129, 5), rebin=75,
-                         chunk_size=1500)
+                         chunk_size=10000)
 print("Got data")
+print("X Shape", x.shape)
+print("Y Shape", y.shape)
 history = ta.Scan(x, y,
                   params=params,
                   dataset_name='separation_test',
                   experiment_no='1',
                   model=input_model,
                   search_method='random',
-                  grid_downsample=0.5)
+                  grid_downsample=0.001)
diff --git a/factnn/utils/cross_validate.py b/factnn/utils/cross_validate.py
@@ -245,7 +245,7 @@ def get_chunk_of_data(directory, proton_directory="", indicies=(30,129,3),
         train_gen, val_gen, test_gen, shape = data(start_slice=indicies[0], end_slice=indicies[1],
                                                    final_slices=indicies[2],
                                                    rebin_size=rebin, gamma_train=gamma_paths,
-                                                   proton_train=proton_paths, batch_size=chunk_size,
+                                                   proton_train=proton_paths, batch_size=1,
                                                    normalize=normalize,
                                                    model_type=model_type, as_channels=as_channels,
                                                    truncate=truncate,
@@ -257,7 +257,7 @@ def get_chunk_of_data(directory, proton_directory="", indicies=(30,129,3),
         train_gen, val_gen, test_gen, shape = data(start_slice=indicies[0], end_slice=indicies[1],
                                                    final_slices=indicies[2],
                                                    rebin_size=rebin, gamma_train=gamma_paths,
-                                                   batch_size=chunk_size, normalize=normalize,
+                                                   batch_size=1, normalize=normalize,
                                                    model_type=model_type, as_channels=as_channels,
                                                    truncate=truncate,
                                                    dynamic_resize=dynamic_resize,
@@ -266,6 +266,10 @@ def get_chunk_of_data(directory, proton_directory="", indicies=(30,129,3),
                                                    return_features=return_features)
 
     x, y = train_gen.__getitem__(0)
+    for i in range(1, chunk_size):
+        x_temp, y_temp = train_gen.__getitem__(i)
+        x = np.append(x, x_temp, axis=0)
+        y = np.append(y, y_temp, axis=0)
 
     return x, y