def test_peak_finding(h5_main):
row_ind, col_ind = 110, 25
pixel_ind = col_ind + row_ind * num_cols
spectra = h5_main[pixel_ind]
peak_inds = find_all_peaks(spectra, [20, 60], num_steps=30)
fig, axis = plt.subplots()
axis.scatter(np.arange(len(spectra)), np.abs(spectra), c='black')
axis.axvline(peak_inds[0], color='r', linewidth=2)
axis.set_ylim([0, 1.1 * np.max(np.abs(spectra))])
axis.set_title('find_all_peaks found peaks at index: {}'.format(peak_inds),
fontsize=16)
def _map_function(spectra, *args, **kwargs):
"""
This is the function that will be applied to each pixel in the dataset.
It's job is to demonstrate what needs to be done for each pixel in the dataset.
pyUSID.Process will handle the parallel computation and memory management
As in typical scientific problems, the results from find_all_peaks() need to be
post-processed
In this case, the find_all_peaks() function can sometimes return 0 or more than one peak
for spectra that are very noisy
Knowing that the peak is typically at the center of the spectra,
we return the central index when no peaks were found
Or the index closest to the center when multiple peaks are found
Finally once we have a single index, we need to index the spectra by that index
in order to get the amplitude at that frequency.
"""
peak_inds = find_all_peaks(spectra, [20, 60], num_steps=30)
central_ind = len(spectra) // 2
if len(peak_inds) == 0:
# too few peaks
# set peak to center of spectra
val = central_ind
elif len(peak_inds) > 1:
# too many peaks
# set to peak closest to center of spectra
dist = np.abs(peak_inds - central_ind)
val = peak_inds[np.argmin(dist)]
else:
# normal situation
val = peak_inds[0]
# Finally take the amplitude of the spectra at this index
return np.abs(spectra[val])
def run_serial_compute(h5_main):
raw_data = h5_main[()]
serial_results = list()
for vector in raw_data:
serial_results.append(find_all_peaks(vector, [20, 60], num_steps=30))
return serial_results
# # The below numpy array is used to configure the returned function wpeaks
# wavelet_widths = np.linspace(width_bounds[0], width_bounds[1], num_steps)
#
# peak_indices = find_peaks_cwt(np.abs(vector), wavelet_widths, **kwargs)
#
# return peak_indices
#
# Testing the function
# ----------------------
# Let’s see what the operation on an example spectra returns.
row_ind, col_ind = 103, 19
pixel_ind = col_ind + row_ind * num_cols
spectra = h5_main[pixel_ind]
peak_inds = find_all_peaks(spectra, [20, 60], num_steps=30)
fig, axis = plt.subplots()
axis.scatter(np.arange(len(spectra)), np.abs(spectra), c='black')
axis.axvline(peak_inds[0], color='r', linewidth=2)
axis.set_ylim([0, 1.1 * np.max(np.abs(spectra))])
axis.set_title('find_all_peaks found peaks at index: {}'.format(peak_inds),
fontsize=16)
########################################################################################################################
# Before we apply the function to the entire dataset, lets load the dataset to memory so that file-loading time is not a
# factor when comparing the times for serial and parallel computing times:
raw_data = h5_main[()]
########################################################################################################################