def pca_reconstruction(self, dataset=None, threshold=None):
"""
Does PCA projection on the normalized noise dataset in order to reconstruct the original
dataset with artifacts removed
TODO: proper threshold that is not based on percentage
:param dataset:
:param threshold: The threshold where the PCA projector will reject principal components
:return:
"""
if dataset is None:
dataset = self.window.clone()
self.normalizer = Normalizer(dataset)
normalized_noised_window = self.normalizer.subtract_means(dataset)
reconstructed_dataset, avg_eigenvalue, max_eigenvalue, rejected = PCA.project(normalized_noised_window, threshold)
reconstructed_window = self.normalizer.add_means(reconstructed_dataset)
return reconstructed_window, avg_eigenvalue, max_eigenvalue, rejected
class Artificer:
def __init__(self, dataset_window):
# self.factors = [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 3, 4, 40, 100, 300, 800, 400, 50, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] # np.linspace(1, 80, len(data[0]))
self.factors = [0, 0, 0, 0, 1, 1, 1, 1, 3, 4, 40, 80, 100, 800]
#self.factors = [0, 0, 0, 0, 0, 0, 0, 0,0, 0, 0, 0, 0, 150]
self.factors = map(lambda x: x*4, self.factors)
self.window = dataset_window
self.spike_range_start = None
self.spike_range_end = None
self.normalizer = None
def add_artifacts(self, spike_size=30):
"""
Adds k noisy artifacts to self.
:param k:
:return:
"""
spike_offset = (len(self.window) - spike_size) // 2
self.spike_range_start = spike_offset
self.spike_range_end = spike_size + spike_offset
data = np.array(self.window.unpack_params())
data_transposed = data.T
mean = np.mean(data_transposed, axis=tuple(range(1, data_transposed.ndim)))
var = np.var(data_transposed, axis=tuple(range(1, data_transposed.ndim)))
self.sine_artifact(self.spike_range_start, self.spike_range_end, data, mean, var)
noise_dataset = DataSet(data.tolist())
return noise_dataset
def sine_artifact(self, spike_range_start, spike_range_end, data, mean, var):
sinus = []
for t in range(spike_range_start, spike_range_end):
d = np.sin((2 * np.pi / (spike_range_end - spike_range_start)) * (t - spike_range_start))
for position in range(len(data[t])):
data[t][position] = data[t][position] + d * self.factors[position]
def pca_reconstruction(self, dataset=None, threshold=None):
"""
Does PCA projection on the normalized noise dataset in order to reconstruct the original
dataset with artifacts removed
TODO: proper threshold that is not based on percentage
:param dataset:
:param threshold: The threshold where the PCA projector will reject principal components
:return:
"""
if dataset is None:
dataset = self.window.clone()
self.normalizer = Normalizer(dataset)
normalized_noised_window = self.normalizer.subtract_means(dataset)
reconstructed_dataset, avg_eigenvalue, max_eigenvalue, rejected = PCA.project(normalized_noised_window, threshold)
reconstructed_window = self.normalizer.add_means(reconstructed_dataset)
return reconstructed_window, avg_eigenvalue, max_eigenvalue, rejected
def mse(self):
original_data = np.array(self.window.unpack_params())
reconstructed_data = np.array(self.reconstructed_window.unpack_params())
sum_window = 0
components_with_artifacts = 0
for i in range(len(original_data)):
for j in range(len(original_data[i])):
sum_window += np.power(original_data[i][j] - reconstructed_data[i][j], 2)
if self.factors[j] > 0:
components_with_artifacts += 1
if components_with_artifacts == 0:
# no artifacts in this window
pass
# MSE for this window
return np.mean(sum_window)
def get_noised_window(self):
return self.noised_window.clone()
