Merge pull request #40 from smoia/enh/metrics

Improve peak detection utility and optimize metrics

Author: smoia

physioqc/metrics/multimodal.py

@@ -3,10 +3,11 @@
 import warnings
 
 import numpy as np
-import peakdet as pk
+from peakdet import Physio
+from peakdet.operations import peakfind_physio
 from scipy import signal
 
-from .utils import has_peakfind_physio, physio_or_numpy
+from .utils import has_peaks, physio_or_numpy
 
 
 @physio_or_numpy
@@ -42,8 +43,7 @@ def std(signal):
     N-sized array :obj:`numpy.ndarray`
         Standard deviation of signal.
     """
-    std_val = np.std(signal, axis=0)
-    return std_val
+    return np.std(signal, axis=0)
 
 
 @physio_or_numpy
@@ -61,8 +61,7 @@ def mean(signal: np.array):
     N-sized array :obj:`numpy.ndarray`
         Mean of signal.
     """
-    mean_val = np.mean(signal, axis=0)
-    return mean_val
+    return np.mean(signal, axis=0)
 
 
 @physio_or_numpy
@@ -80,8 +79,7 @@ def tSNR(signal):
     N-sized array :obj:`numpy.ndarray`
         Temporal signal to noise ratio of signal.
     """
-    tSNR_val = np.mean(signal, axis=0) / np.std(signal, axis=0)
-    return tSNR_val
+    return np.mean(signal, axis=0) / np.std(signal, axis=0)
 
 
 @physio_or_numpy
@@ -99,8 +97,7 @@ def CoV(signal):
     N-sized array :obj:`numpy.ndarray`
         Temporal signal to noise ratio of signal.
     """
-    cov_val = np.std(signal, axis=0) / np.mean(signal, axis=0)
-    return cov_val
+    return np.std(signal, axis=0) / np.mean(signal, axis=0)
 
 
 @physio_or_numpy
@@ -118,8 +115,7 @@ def min(signal: np.array):
     N-sized array :obj:`numpy.ndarray`
         min of signal.
     """
-    min_val = np.min(signal, axis=0)
-    return min_val
+    return np.min(signal, axis=0)
 
 
 @physio_or_numpy
@@ -137,8 +133,7 @@ def max(signal: np.array):
     N-sized array :obj:`numpy.ndarray`
         max of signal.
     """
-    max_val = np.max(signal, axis=0)
-    return max_val
+    return np.max(signal, axis=0)
 
 
 @physio_or_numpy
@@ -160,9 +155,8 @@ def iqr(signal: np.array, q_high: float = 75, q_low: float = 25):
         iqr of the signal
     """
     p_high, p_low = np.percentile(signal, [q_high, q_low], axis=0)
-    iqr_val = p_high - p_low
 
-    return iqr_val
+    return p_high - p_low
 
 
 @physio_or_numpy
@@ -181,13 +175,11 @@ def percentile(signal: np.array, perc: float = 2):
     np.array
         percentile of the signal
     """
-    perc_val = np.percentile(signal, axis=0, q=perc)
-
-    return perc_val
+    return np.percentile(signal, q=perc, axis=0)
 
 
 def peak_detection(
-    data: pk.Physio,
+    data: Physio,
     peak_threshold: float = 0.1,
     peak_dist: float = 60.0,
 ):
@@ -196,7 +188,7 @@ def peak_detection(
 
     Parameters
     ----------
-    data : pk.Physio
+    data : Physio
         A peakdet Physio object
     peak_threshold : float, optional
         Threshold for peak detection, by default 0.1
@@ -205,58 +197,60 @@ def peak_detection(
 
     Returns
     -------
-    pk.Physio
-        Updated pk.Physio class with peaks etc.
+    Physio
+        Updated Physio class with peaks etc.
     """
-    ph = pk.operations.peakfind_physio(data, thresh=peak_threshold, dist=peak_dist)
-
-    return ph
+    return peakfind_physio(data, thresh=peak_threshold, dist=peak_dist)
 
 
-def peak_distance(ph: pk.Physio):
-    """Calculate the time between peaks (first derivative of onsets).
+def peak_distance(
+    ph: Physio,
+    peak_threshold: float = 0.1,
+    peak_dist: float = 60.0,
+):
+    """Calculate the time (in seconds) between peaks (first derivative of onsets).
 
     Parameters
     ----------
-    ph : pk.Physio
-        A pk.Physio object, that contains peak information.
+    ph : Physio
+        A Physio object, that contains peak information.
 
     Returns
     -------
     np.array
         np.array of shape [npeaks, ]
     """
-    if not has_peakfind_physio(ph):
+    if not has_peaks(ph):
         warnings.warn("Peaks not estimated, estimating")
-        ph = peak_detection(ph)
-
-    diff_peak = np.diff(ph.peaks, axis=0)
+        ph = peakfind_physio(ph, thresh=peak_threshold, dist=peak_dist)
 
-    return diff_peak
+    return np.diff(ph.peaks, axis=0) / ph.fs
 
 
-def peak_amplitude(ph: pk.Physio):
+def peak_amplitude(
+    ph: Physio,
+    peak_threshold: float = 0.1,
+    peak_dist: float = 60.0,
+):
     """Return the amplitude for each peak in the ph.Physio object (peak - trough).
 
     Parameters
     ----------
-    ph : pk.Physio
-        pk.Physio object with peak and trough information.
+    ph : Physio
+        Physio object with peak and trough information.
 
     Returns
     -------
     np.array
         np.array of shape [npeaks - 1, ]
     """
-    if not has_peakfind_physio(ph):
+    if not has_peaks(ph):
         warnings.warn("Peaks not estimated, estimating")
-        ph = peak_detection(ph)
+        ph = peakfind_physio(ph, thresh=peak_threshold, dist=peak_dist)
     # Assuming that peaks and troughs are aligned. Last peak has no trough.
     peak_amp = ph.data[ph.peaks[:-1]]
     trough_amp = ph.data[ph.troughs]
-    peak_amplitude = peak_amp - trough_amp
-
-    return peak_amplitude
+    return peak_amp - trough_amp
 
 
 def power_spectrum(data):
@@ -309,9 +303,7 @@ def energy(data, lowf=None, highf=None):
         # Define frequency band
         idx_band = np.logical_and(freqs >= lowf, freqs <= highf)
 
-    energy = np.sum(energy_density[idx_band])
-
-    return energy
+    return np.sum(energy_density[idx_band])
 
 
 def fALFF(data, lowf=0, highf=0.5):
@@ -346,9 +338,7 @@ def fALFF(data, lowf=0, highf=0.5):
     total_energy = energy(data)
 
     # Compute the relative energy
-    rel_energy = band_energy / total_energy
-
-    return rel_energy
+    return band_energy / total_energy
 
 
 def freq_energy(data, thr=0.5):
@@ -372,6 +362,4 @@ def freq_energy(data, thr=0.5):
     The value of the threshold has been selected randomly for now. Please update it with a more meaningful value.
     """
     energy_nd = energy(data)
-    freq = np.argmax(energy_nd > thr)
-
-    return freq
+    return np.argmax(energy_nd > thr)

physioqc/metrics/utils.py

@@ -3,6 +3,8 @@
 import functools
 import logging
 
+from peakdet.physio import Physio
+
 LGR = logging.getLogger(__name__)
 LGR.setLevel(logging.INFO)
 
@@ -58,8 +60,8 @@ def wrapper(*args, **kwargs):
     return wrapper
 
 
-def has_peakfind_physio(signal) -> bool:
-    """Check if "peakfind_physio" is in signal's history.
+def has_peaks(signal: Physio) -> bool:
+    """Check if the signal is a Physio object and has (at least 2) peaks.
 
     Parameters
     ----------
@@ -77,8 +79,9 @@ def has_peakfind_physio(signal) -> bool:
         Raises error if object does not have a history attribute.
     """
     if not hasattr(signal, "history"):
-        raise AttributeError("Signal has to be a Physio object!")
+        raise AttributeError("Signal has to be a peakdet Physio object!")
 
-    has_peakfind = any(["peakfind_physio" in i for i in signal.history])
+    if signal._metadata["peaks"].size == 1:
+        LGR.warn("Signal has only one peak! Better to rerun peak detection.")
 
-    return has_peakfind
+    return signal._metadata["peaks"].size > 1