def test_save_and_load_model(self):
# Load and train the model
self.classifier.load()
# Load samples multiple times to have a large dataset
for n in range(100):
self.classifier.load_samples_data()
self.assertGreater(self.classifier.train_model(), 0.9)
model_path = os.path.join("test_dir_svm_classifier", "model.svm")
# Save the model
self.classifier.save_model(model_path)
# Create a new classifier and load the saved model
new_classifier = SVMClassifier(model_path=model_path)
# Load the model from the saved model
new_classifier.load()
# Check if the loaded model works correctly
test_sample1 = Sample(data=[[1], [1]])
test_sample2 = Sample(data=[[1], [0]])
self.assertEqual(new_classifier.predict(test_sample1), "0")
self.assertEqual(new_classifier.predict(test_sample2), "1")
def run_example(*args, **kwargs):
# Create the SerialDataReader
sdr = SerialDataReader(kwargs['port'], expected_axis=6, verbose=False)
# Create the SampleManager
manager = DiscreteSampleManager()
# Attach the manager
sdr.attach_manager(manager)
# Create a classifier
classifier = SVMClassifier(model_path=args[0])
# Load the model
classifier.load()
# Create a ClassifierPredictor
predictor = ClassifierPredictor(classifier)
# Attach the classifier predictor
manager.attach_receiver(predictor)
# Create a CallbackManager
callback_mg = CallbackManager(verbose=True)
# Attach the callback manager
predictor.attach_callback_manager(callback_mg)
# Open the serial connection
sdr.open()
print("Opened!")
# Start the main loop
sdr.mainloop()
def train_svm_classifier(dataset_dir, output_file, n_jobs=1):
"""
Train an SVM model from the given dataset and save it to a file.
"""
# Create the classifier
classifier = SVMClassifier(dataset_path=dataset_dir, verbose=True, n_jobs=n_jobs,
autoscale_size=50)
# Train the classifier
train_classifier(classifier=classifier, dataset_dir=dataset_dir, output_file=output_file, n_jobs=n_jobs)
def setUp(self):
# Create a test directory if it doesn't exists
if not os.path.exists("test_dir_svm_classifier"):
os.makedirs("test_dir_svm_classifier")
# Create the samples with the 4 possibilities of the XOR gate
sample1 = Sample(data=[[1], [1]], gesture_id="0")
sample2 = Sample(data=[[0], [0]], gesture_id="0")
sample3 = Sample(data=[[1], [0]], gesture_id="1")
sample4 = Sample(data=[[0], [1]], gesture_id="1")
# Save the samples in the test directory
sample1.save_to_file(os.path.join("test_dir_svm_classifier", "0_1.txt"))
sample2.save_to_file(os.path.join("test_dir_svm_classifier", "0_2.txt"))
sample3.save_to_file(os.path.join("test_dir_svm_classifier", "1_1.txt"))
sample4.save_to_file(os.path.join("test_dir_svm_classifier", "1_2.txt"))
# Initialize an AbstractClassifier
self.classifier = SVMClassifier(dataset_path="test_dir_svm_classifier", test_size=0.5)
def run_example(*args, **kwargs):
# Create the SerialDataReader
sdr = SerialDataReader(kwargs['port'], expected_axis=6, verbose=False)
# Create the SampleManager
manager = StreamSampleManager(step=20, window=20)
# Attach the manager
sdr.attach_manager(manager)
# Create a threshold middleware
middleware = GradientThresholdMiddleware(verbose=False,
threshold=40,
sample_group_delay=5,
group=True)
# Attach the middleware
manager.attach_receiver(middleware)
# Create a classifier
classifier = SVMClassifier(model_path=args[0])
# Load the model
classifier.load()
# Print classifier info
classifier.print_info()
# Create a ClassifierPredictor
predictor = ClassifierPredictor(classifier)
# Filter the samples that are too short or too long
lfmiddleware = LengthThresholdMiddleware(verbose=True,
min_len=180,
max_len=600)
middleware.attach_receiver(lfmiddleware)
# Attach the classifier predictor
lfmiddleware.attach_receiver(predictor)
# Create a CallbackManager
callback_mg = CallbackManager(verbose=True)
# Attach the callback manager
predictor.attach_callback_manager(callback_mg)
# Attach the callbacks
callback_mg.attach_callback("knock", receive_gesture)
callback_mg.attach_callback("doubleknock", receive_gesture)
# Open the serial connection
sdr.open()
print("Opened!")
# Start the main loop
sdr.mainloop()
def train(dataset_dir, output_file, n_jobs=1):
"""
Used to train the knock sensor data.
Train an MLP model from the given dataset and save it to a file.
"""
tm = TrimmerMiddleware(threshold=300)
fft = FFTMiddleware()
# Create the classifier
classifier = SVMClassifier(dataset_path=dataset_dir, verbose=True, n_jobs=n_jobs,
autonormalize=False, autoscale_size=500, middlewares=[tm, fft])
# Train the classifier
train_classifier(classifier=classifier, dataset_dir=dataset_dir, output_file=output_file, n_jobs=n_jobs)
def receive_character(number):
pyautogui.doubleClick(155, 69)
if number != "1":
pyautogui.typewrite(number)
else:
pyautogui.keyDown('backspace')
pyautogui.keyUp('backspace')
# Create the SerialDataReader
sdr = SerialDataReader(PORT, expected_axis=6, verbose=False)
# Create the SampleManager
manager = StreamSampleManager()
# Attach the manager
sdr.attach_manager(manager)
# Create a classifier
classifier = SVMClassifier(model_path=MODEL_PATH)
# Load the model
classifier.load()
# Print classifier info
classifier.print_info()
# Create a ClassifierPredictor
predictor = ClassifierPredictor(classifier)
# Attach the classifier predictor
manager.attach_receiver(predictor)
# Create a CallbackManager
callback_mg = CallbackManager(verbose=True)
# Attach the callback manager
predictor.attach_callback_manager(callback_mg)
# Bind all the numbers
callback_mg.attach_callback(" ", receive_character)
for c in ["1", "2", "3", "4"]:
callback_mg.attach_callback(c, receive_character)
# Open the serial connection
sdr.open()
print("Opened!")
# Start the main loop
sdr.mainloop()
def train(dataset_dir, output_file, n_jobs=1):
"""
Used to train the knock sensor data.
Train an MLP model from the given dataset and save it to a file.
"""
#fftm = FFTMiddleware()
asm = AbsoluteScaleMiddleware(subtract=500, rolling_mean_window=4)
# Create the classifier
classifier = SVMClassifier(dataset_path=dataset_dir,
verbose=True,
n_jobs=n_jobs,
autoscale_size=50,
middlewares=[asm])
# Train the classifier
train_classifier(classifier=classifier,
dataset_dir=dataset_dir,
output_file=output_file,
n_jobs=n_jobs)
def train(dataset_dir, output_file, n_jobs=1):
"""
Used to train the knock sensor data.
Train an MLP model from the given dataset and save it to a file.
"""
asm = AbsoluteScaleMiddleware()
# Create the classifier
classifier = SVMClassifier(dataset_path=dataset_dir,
verbose=True,
n_jobs=n_jobs,
autonormalize=True,
autoscale_size=40,
middlewares=[asm])
# Train the classifier
train_classifier(classifier=classifier,
dataset_dir=dataset_dir,
output_file=output_file,
n_jobs=n_jobs)
def receive_character(number):
print(number)
# Create the SerialDataReader
sdr = SerialDataReader(PORT, expected_axis=6, verbose=False)
# Create the SampleManager
manager = DiscreteSampleManager()
# Attach the manager
sdr.attach_manager(manager)
# Create a classifier
classifier = SVMClassifier(model_path=MODEL_PATH)
# Load the model
classifier.load()
# Print classifier info
classifier.print_info()
# Create a ClassifierPredictor
predictor = ClassifierPredictor(classifier)
# Attach the classifier predictor
manager.attach_receiver(predictor)
# Create a CallbackManager
callback_mg = CallbackManager(verbose=True)
from pygarl.classifiers import SVMClassifier
from pygarl.middlewares import TrimmerMiddleware, FFTMiddleware
from pygarl.plugins.train import train_classifier
import sys
dataset_dir = sys.argv[1]
# If launched directly, parse the parameters from sys
if __name__ == '__main__':
tm = TrimmerMiddleware(threshold=300)
fft = FFTMiddleware()
# Create the classifier
classifier = SVMClassifier(dataset_path=dataset_dir, verbose=True, n_jobs=8,
autonormalize=False, autoscale_size=500, middlewares=[tm, fft])
# Train the classifier
train_classifier(classifier=classifier, dataset_dir=dataset_dir, output_file="model.svm", n_jobs=8)
sdr = SerialDataReader(sys.argv[1], expected_axis=6, verbose=False) # Create the SampleManager manager = StreamSampleManager(step=20, window=20) # Attach the manager sdr.attach_manager(manager) # Create a threshold middleware middleware = GradientThresholdMiddleware(verbose=False, threshold=40, sample_group_delay=5, group=True) # Attach the middleware manager.attach_receiver(middleware) # Create a classifier classifier = SVMClassifier(model_path=sys.argv[2]) # Load the model classifier.load() # Print classifier info classifier.print_info() # Create a ClassifierPredictor predictor = ClassifierPredictor(classifier) # Filter the samples that are too short or too long lfmiddleware = LengthThresholdMiddleware(verbose=False, min_len=180, max_len=600) middleware.attach_receiver(lfmiddleware) # Attach the classifier predictor
class SVMClassifierTestCase(unittest.TestCase):
"""
Tests to check SVMClassifier consistency.
In order to test the classifier, this test will try to implement
a XOR gate using the classifier.
"""
def setUp(self):
# Create a test directory if it doesn't exists
if not os.path.exists("test_dir_svm_classifier"):
os.makedirs("test_dir_svm_classifier")
# Create the samples with the 4 possibilities of the XOR gate
sample1 = Sample(data=[[1], [1]], gesture_id="0")
sample2 = Sample(data=[[0], [0]], gesture_id="0")
sample3 = Sample(data=[[1], [0]], gesture_id="1")
sample4 = Sample(data=[[0], [1]], gesture_id="1")
# Save the samples in the test directory
sample1.save_to_file(os.path.join("test_dir_svm_classifier", "0_1.txt"))
sample2.save_to_file(os.path.join("test_dir_svm_classifier", "0_2.txt"))
sample3.save_to_file(os.path.join("test_dir_svm_classifier", "1_1.txt"))
sample4.save_to_file(os.path.join("test_dir_svm_classifier", "1_2.txt"))
# Initialize an AbstractClassifier
self.classifier = SVMClassifier(dataset_path="test_dir_svm_classifier", test_size=0.5)
def tearDown(self):
# Destroy the test directory
shutil.rmtree("test_dir_svm_classifier")
# Destroy the AbstractClassifier
self.classifier = None
def test_load_samples_loaded_correctly(self):
# Load the samples
self.classifier.load()
# Test if samples has been loaded correctly
# Check there are 2 0s and 2 1s
self.assertEqual(len(list(filter(lambda x: x == 0, self.classifier.y_data))), 2)
self.assertEqual(len(list(filter(lambda x: x == 1, self.classifier.y_data))), 2)
# Convert to list from numpy array
x_data_as_list = map(lambda x: x.tolist(), self.classifier.x_data)
self.assertIn([1, 1], x_data_as_list)
self.assertIn([1, 0], x_data_as_list)
self.assertIn([0, 1], x_data_as_list)
self.assertIn([0, 0], x_data_as_list)
# Test if gestures ids has been loaded correctly
self.assertIn("0", self.classifier.gestures)
self.assertIn("1", self.classifier.gestures)
def test_train_model(self):
self.classifier.load()
# Load samples multiple times to have a large dataset
for n in range(100):
self.classifier.load_samples_data()
self.assertGreater(self.classifier.train_model(), 0.9)
def test_predict(self):
self.classifier.load()
# Load samples multiple times to have a large dataset
for n in range(100):
self.classifier.load_samples_data()
self.assertGreater(self.classifier.train_model(), 0.9)
test_sample1 = Sample(data=[[1], [1]])
test_sample2 = Sample(data=[[1], [0]])
self.assertEqual(self.classifier.predict(test_sample1), "0")
self.assertEqual(self.classifier.predict(test_sample2), "1")
def test_save_model_before_training_should_fail(self):
self.assertRaises(ValueError, self.classifier.save_model, "path")
def test_save_and_load_model(self):
# Load and train the model
self.classifier.load()
# Load samples multiple times to have a large dataset
for n in range(100):
self.classifier.load_samples_data()
self.assertGreater(self.classifier.train_model(), 0.9)
model_path = os.path.join("test_dir_svm_classifier", "model.svm")
# Save the model
self.classifier.save_model(model_path)
# Create a new classifier and load the saved model
new_classifier = SVMClassifier(model_path=model_path)
# Load the model from the saved model
new_classifier.load()
# Check if the loaded model works correctly
test_sample1 = Sample(data=[[1], [1]])
test_sample2 = Sample(data=[[1], [0]])
self.assertEqual(new_classifier.predict(test_sample1), "0")
self.assertEqual(new_classifier.predict(test_sample2), "1")
