def apply(self, data, meta=None):
all_ceps = []
for ch in data:
ceps, mspec, spec = mfcc(ch)
all_ceps.append(ceps.ravel())
return to_np_array(all_ceps)
def apply(self, data, meta=None):
all_ceps = []
for ch in data:
ceps, mspec, spec = mfcc(ch)
all_ceps.append(ceps.ravel())
return to_np_array(all_ceps)
def apply_one(self, fft_mag, meta):
num_time_samples = (fft_mag.shape[-1] -
1) * 2 # revert FFT shape change
X = fft_mag**2
for ch in X:
ch /= np.sum(ch + 1e-12)
psd = X # pdf
out = []
#[0,1,2] -> [[0,1], [1,2]]
for start_freq, end_freq in zip(self.freq_ranges[:-1],
self.freq_ranges[1:]):
start_index = np.floor(
(start_freq / meta.sampling_frequency) * num_time_samples)
end_index = np.floor(
(end_freq / meta.sampling_frequency) * num_time_samples)
selected = psd[:, start_index:end_index]
entropies = -np.sum(selected * np.log2(selected + 1e-12),
axis=selected.ndim - 1) / np.log2(end_index -
start_index)
if self.flatten:
out.append(entropies.ravel())
else:
out.append(entropies)
if self.flatten:
return np.concatenate(out)
else:
return to_np_array(out)
def upper_right_triangle(matrix):
accum = []
for i in range(matrix.shape[0]):
for j in range(i + 1, matrix.shape[1]):
accum.append(matrix[i, j])
return to_np_array(accum)
def upper_right_triangle(matrix):
accum = []
for i in range(matrix.shape[0]):
for j in range(i+1, matrix.shape[1]):
accum.append(matrix[i, j])
return to_np_array(accum)
def apply_one(self, fft_mag, meta):
num_time_samples = (fft_mag.shape[-1] - 1) * 2 # revert FFT shape change
X = fft_mag ** 2
for ch in X:
ch /= np.sum(ch + 1e-12)
psd = X # pdf
out = []
#[0,1,2] -> [[0,1], [1,2]]
for start_freq, end_freq in zip(self.freq_ranges[:-1], self.freq_ranges[1:]):
start_index = np.floor((start_freq / meta.sampling_frequency) * num_time_samples)
end_index = np.floor((end_freq / meta.sampling_frequency) * num_time_samples)
selected = psd[:, start_index:end_index]
entropies = - np.sum(selected * np.log2(selected + 1e-12), axis=selected.ndim-1) / np.log2(end_index - start_index)
if self.flatten:
out.append(entropies.ravel())
else:
out.append(entropies)
if self.flatten:
return np.concatenate(out)
else:
return to_np_array(out)
def apply(self, datas, meta):
if datas.ndim >= 3:
out = []
for d in datas:
out.append(self.apply_one(d, meta))
return to_np_array(out)
else:
return self.apply_one(datas, meta)
def apply(self, datas, meta):
if datas.ndim >= 3:
out = []
for d in datas:
out.append(self.apply_one(d, meta))
return to_np_array(out)
else:
return self.apply_one(datas, meta)
def apply(self, X, meta=None):
if self.window_secs is None:
return X.reshape([1] + list(X.shape))
num_windows = meta.data_length_sec / self.window_secs
samples_per_window = self.window_secs * int(meta.sampling_frequency)
samples_used = num_windows * samples_per_window
samples_dropped = X.shape[-1] - samples_used
X = Slice(samples_dropped).apply(X)
out = np.split(X, num_windows, axis=X.ndim - 1)
out = to_np_array(out)
return out
def apply(self, X, meta=None):
if self.window_secs is None:
return X.reshape([1] + list(X.shape))
num_windows = meta.data_length_sec / self.window_secs
samples_per_window = self.window_secs * int(meta.sampling_frequency)
samples_used = num_windows * samples_per_window
samples_dropped = X.shape[-1] - samples_used
X = Slice(samples_dropped).apply(X)
out = np.split(X, num_windows, axis=X.ndim-1)
out = to_np_array(out)
return out
def apply(self, X, meta):
if meta.data_type == 'preictal':
num_windows = (meta.data_length_sec / self.window_secs) * self.gen_factor
samples_per_window = self.window_secs * int(meta.sampling_frequency) / self.gen_factor
samples_used = num_windows * samples_per_window
samples_dropped = X.shape[-1] - samples_used
X = Slice(samples_dropped).apply(X)
pieces = np.split(X, num_windows, axis=X.ndim-1)
pieces_per_window = self.gen_factor
gen = [np.concatenate(pieces[i:i+pieces_per_window], axis=pieces[0].ndim - 1) for i in range(0, num_windows - self.gen_factor + 1)]
gen = to_np_array(gen)
return gen
else:
return self.windower.apply(X, meta)
def apply(self, X, meta):
if meta.data_type == 'preictal':
num_windows = (meta.data_length_sec /
self.window_secs) * self.gen_factor
samples_per_window = self.window_secs * int(
meta.sampling_frequency) / self.gen_factor
samples_used = num_windows * samples_per_window
samples_dropped = X.shape[-1] - samples_used
X = Slice(samples_dropped).apply(X)
pieces = np.split(X, num_windows, axis=X.ndim - 1)
pieces_per_window = self.gen_factor
gen = [
np.concatenate(pieces[i:i + pieces_per_window],
axis=pieces[0].ndim - 1)
for i in range(0, num_windows - self.gen_factor + 1)
]
gen = to_np_array(gen)
return gen
else:
return self.windower.apply(X, meta)
def apply_one(self, data, meta=None):
return to_np_array([hfd(ch, self.kmax) for ch in data])
def apply_one(self, X, meta=None):
return to_np_array([self.pfd_for_ch(ch) for ch in X])
def apply_one(self, data, meta=None):
return to_np_array([hfd(ch, self.kmax) for ch in data])
def apply_one(self, X, meta=None):
return to_np_array([self.pfd_for_ch(ch) for ch in X])
