Select path for training set

sensorcnn/tests/setup_mlp_test.py

@@ -8,6 +8,7 @@
 from dataset.utils import *
 from train.utils import *
 
+
 def test_mlp_1():
   model = MuvrSequential()
   #TODO: pass the name to the constructor
@@ -30,17 +31,39 @@ def test_mlp_2():
   train_mlp(model, nb_epoch=100)
 
 
-def train_mlp(model, optimizer=None, nb_epoch=30):
-  print("Load all samples")
+def test_mlp1_steve():
+  model = MuvrSequential()
+  model.name = "mlp_1_steve"
+  model.layers_as_string = "750 id 150 tanh 100 tanh 2 tanh"
+  model.add(Dense(input_dim=750, output_dim=150, activation="tanh", init='uniform'))
+  model.add(Dropout(0.5))
+  model.add(Dense(input_dim=150, output_dim=100, activation="tanh", init='uniform'))
+  model.add(Dropout(0.5))
+  model.add(Dense(input_dim=100, output_dim=2, activation="tanh", init='uniform'))
+  setup_labels = ["setup_resistanceTargeted:arms/dumbbell-biceps-curl", "setup_resistanceTargeted:arms/triceps-extension"]
+  train_mlp(model, nb_epoch=100, data_path="/data/Setup Data/Steve/", setup_labels=setup_labels)
+
+
+def test_mlp2_steve():
+  model = MuvrSequential()
+  model.name = "mlp_2_steve"
+  model.layers_as_string = "750 id 2 tanh"
+  model.add(Dense(input_dim=750, output_dim=2, activation="tanh", init='uniform'))
+  setup_labels = ["setup_resistanceTargeted:arms/dumbbell-biceps-curl", "setup_resistanceTargeted:arms/triceps-extension"]
+  train_mlp(model, nb_epoch=100, data_path="/data/Setup Data/Steve/", setup_labels=setup_labels)
+
+
+def train_mlp(model, optimizer=None, nb_epoch=30, data_path="/data/Setup Data/", setup_labels=None):
   all_samples = {}
-  for filename in csv_file_iterator("/data/Setup Data/"):
+  for filename in csv_file_iterator(data_path):
       samples = load_from_csv(filename)
       all_samples[filename] = samples
 
   from functools import reduce
   df = reduce(lambda x, y: x.append(y), all_samples.values())
+  if not setup_labels:
+    setup_labels = [l for l in df['label'].value_counts().keys() if l.startswith("setup_")]
 
-  setup_labels = [l for l in df['label'].value_counts().keys() if l.startswith("setup_")]
   setup_labels_map = {l:i+1 for (i,l) in enumerate(setup_labels)}
   label_to_idx = lambda l: setup_labels_map.get(l, 0)
 
@@ -49,43 +72,35 @@ def train_mlp(model, optimizer=None, nb_epoch=30):
   targets = positive_targets[['target']].as_matrix()
   samples = df[['x', 'y', 'z']].as_matrix()[positive_targets.index]
 
-  print("Shapes before augmentation")
-  print(samples.shape)
-  print(targets.shape)
-
   X, y = augment_examples(samples, targets, new_sample_size=50*5)
 
-  print("Shapes after augmentation")
-  print(X.shape)
-  print(y.shape)
-
-  Y = np.zeros((y.shape[0], 4))
+  output_dim = len(setup_labels)
+  Y = np.zeros((y.shape[0], output_dim))
   iy = np.array([[i, yi-1] for (i, yi) in enumerate(y) if yi > 0])
   Y[iy[:, 0], iy[:, 1]] = 1
 
   # Split Data
   X_train, X_test, Y_train, Y_test = split(X, Y)
 
-  print("Build the model")
   if not optimizer:
     optimizer = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
 
   model.compile(loss='mean_squared_error', optimizer=optimizer, metrics=["accuracy"])
 
-  model.fit(X_train, Y_train, nb_epoch=nb_epoch, batch_size=128)
+  model.fit(X_train, Y_train, nb_epoch=nb_epoch, batch_size=128, verbose=0)
 
   # Printing the results
-  train_score = model.evaluate(X_train, Y_train, batch_size=128)
-  test_score = model.evaluate(X_test, Y_test, batch_size=128)
-  print(train_score)
-  print(test_score)
+  train_score = model.evaluate(X_train, Y_train, batch_size=128, verbose=0)
+  test_score = model.evaluate(X_test, Y_test, batch_size=128, verbose=0)
+  print("\ntrain score\tcost: %s\taccuracy: %s" % tuple(train_score))
+  print("test score\tcost: %s\taccuracy: %s" % tuple(test_score))
 
   # Export result into csv file
   y_true = np.array([np.argmax(yi) + 1 if np.sum(yi) > 0 else 0 for yi in Y_test])
   y_pred = np.array([np.argmax(yi) + 1 if np.sum(yi) > 0 else 0 for yi in model.predict(X_test)])
   evaluation_tabel = evaluate(y_true, y_pred, setup_labels)
 
   # Continue testing on the test set and export the model
-  model.fit(X_test, Y_test, nb_epoch=nb_epoch, batch_size=128)
+  model.fit(X_test, Y_test, nb_epoch=nb_epoch, batch_size=128, verbose=0)
   labels = [l.split("setup_")[-1] for l in setup_labels]
   export(model, evaluation_tabel, labels)