Merge pull request #131 from lifewatch/feature/remove_nmf

removed nmf

pypam/acoustic_file.py

@@ -18,7 +18,6 @@
 import xarray
 from tqdm.auto import tqdm
 
-from pypam import nmf
 from pypam import plots
 from pypam import signal as sig
 from pypam import utils
@@ -980,41 +979,6 @@ def spd(self, binsize=None, bin_overlap=0, h=0.1, nfft=512, fft_overlap=0.5,
                                  bin_overlap=bin_overlap, band=band)
         return utils.compute_spd(psd_evolution, h=h, percentiles=percentiles, max_val=max_val, min_val=min_val)
 
-    def source_separation(self, window_time=1.0, n_sources=15, binsize=None, save_path=None, verbose=False, band=None):
-        """
-        Perform non-negative Matrix Factorization to separate sources
-
-        Parameters
-        ----------
-        window_time: float
-            window time to consider in seconds
-        n_sources : int
-            Number of sources
-        binsize : float
-            Time window considered, in seconds. If set to None, only one value is returned
-        save_path: str or Path
-            Where to save the output
-        verbose: bool
-            Set to True to make plots of the process
-        band : tuple or None
-            Band to filter the spectrogram in. A band is represented with a tuple - or a list - as
-            (low_freq, high_freq). If set to None, the broadband up to the Nyquist frequency will be analyzed
-
-        """
-        if band is None:
-            band = [None, self.fs / 2]
-        separator = nmf.NMF(window_time=window_time, rank=n_sources, save_path=save_path)
-        ds = xarray.Dataset()
-        for i, time_bin, signal, start_sample, end_sample in self._bins(binsize, bin_overlap=0.0):
-            signal.set_band(band)
-            separation_ds = separator(signal, verbose=verbose)
-            separation_ds = separation_ds.assign_coords({'id': [i], 'datetime': ('id', [time_bin])})
-            if i == 0:
-                ds = separation_ds
-            else:
-                ds = xarray.concat((ds, separation_ds), 'id')
-        return ds
-
     def plot_spectrum_median(self, scaling='density', db=True, log=True, save_path=None, **kwargs):
         """
         Plot the power spectrogram density of all the file (units^2 / Hz) re 1 V 1 upa

pypam/acoustic_survey.py

@@ -352,30 +352,6 @@ def hybrid_millidecade_bands(self, db=True, method='spectrum', band=None, percen
         spectra_ds['millidecade_bands'] = milli_spectra
         return spectra_ds
 
-    def source_separation(self, window_time=1.0, n_sources=15, save_path=None, verbose=False, band=None):
-        """
-        Separate the signal in n_sources sources, using non-negative matrix factorization
-        Parameters
-        ----------
-        window_time: float
-            Duration of the window in seconds
-        n_sources: int
-            Number of sources to separate the sound in
-        save_path: str or Path
-            Where to save the output
-        verbose: bool
-            Set to True to make plots of the process
-        band : tuple or list
-            Tuple or list with two elements: low-cut and high-cut of the band to analyze
-        """
-        ds = xarray.Dataset(attrs=self._get_metadata_attrs())
-        for sound_file in self._files():
-            nmf_ds = sound_file.source_separation(window_time, n_sources, binsize=self.binsize, band=band,
-                                                  save_path=save_path, verbose=verbose)
-            ds = utils.merge_ds(ds, nmf_ds, self.file_dependent_attrs)
-
-        return ds
-
     def plot_rms_evolution(self, db=True, save_path=None):
         """
         Plot the rms evolution

tests/test_acoustic_file.py

@@ -34,10 +34,6 @@ def test_millidecade_bands(self):
         self.acu_file.hybrid_millidecade_bands(nfft, fft_overlap=0.5, binsize=None, bin_overlap=0, db=True,
                                                method='density', band=None)
 
-    def test_nmf(self):
-        self.acu_file.source_separation(window_time=1.0, n_sources=15,
-                                        binsize=None, save_path=None, verbose=False, band=None)
-
     def test_update_freq_cal(self):
         ds_psd = self.acu_file.psd()
         ds_psd_updated = self.acu_file.update_freq_cal(ds=ds_psd, data_var='band_density')

tests/test_acoustic_survey.py

@@ -65,9 +65,6 @@ def test_path_not_exists(self):
     def test_timestamp_array(self):
         self.asa.timestamps_array()
 
-    def test_nmf(self):
-        ds = self.asa.source_separation(window_time=1.0, n_sources=15, save_path=None, verbose=verbose)
-
     def test_features(self):
         self.asa.evolution_multiple(method_list=fast_features, band_list=band_list)
         ds = self.asa.evolution_multiple(method_list=acoustic_indices_features, min_freq=0, max_freq=4000,

tests/test_signal.py

@@ -1,6 +1,5 @@
 import unittest
 import pypam.signal as sig
-import pypam.nmf as nmf
 import numpy as np
 from tests import skip_unless_with_plots, with_plots
 import matplotlib.pyplot as plt
@@ -35,14 +34,6 @@ def test_spectrum(self):
         plt.plot(fbands, spectra)
         plt.show()
 
-    def test_source_separation(self):
-        separator = nmf.NMF(window_time=0.1, rank=15)
-        s = sig.Signal(self.data, fs=fs)
-        s.set_band(None)
-        separation_ds = separator(s, verbose=with_plots())
-        reconstructed_sources = separator.reconstruct_sources(separation_ds)
-        separator.return_filtered_signal(s, reconstructed_sources['C_tf'])
-
     def test_acoustic_indices(self):
         s = sig.Signal(self.data, fs=fs)
         aci = s.aci()