README; docs; better references; icwt arg fix

CHANGELOG.md

@@ -15,6 +15,7 @@
  - `istft` edge case with `dtype='float32'`, time-localized `window`, and large `hop_len`
  - `istft` default `N` was supposed to be "largest possible" but wasn't; fixed
  - `istft` division by window norm now accounts for values below float `tiny`
+ - `icwt` defaults args fix
  - fixed Ridge Extraction README
  - fixed important typo in `ridge_extraction.py`
  - fixed `nv` criterion for ignoring sanity check
@@ -29,6 +30,8 @@
  - simplify `p2up`
  - `nan_checks` now defaults to `True` only for NumPy inputs
  - CI changes: conda -> micromamba
+ - docstring references improved
+ - README section added
 
 
 ### 0.6.3 (1-23-2022): QoL, cleanups, fixes

README.md

@@ -169,6 +169,13 @@ imshow(Sx, **ikw, yticks=freqs_stft)
 Also see ridge extraction [README](https://github.com/OverLordGoldDragon/ssqueezepy/tree/master/examples/ridge_extraction).
 
 
+## Interesting use cases (with code)
+
+ 1. [Identify abrupt changes in audio](https://dsp.stackexchange.com/a/87512/50076) - `ssq_cwt` and `ssq_stft` used together to solve an ML problem without ML
+
+Feel free to share yours [here](https://github.com/OverLordGoldDragon/ssqueezepy/issues/9).
+
+
 ## Learning resources
 
  1. [Continuous Wavelet Transform, & vs STFT](https://ccrma.stanford.edu/~unjung/mylec/WTpart1.html)

ssqueezepy/_cwt.py

@@ -72,8 +72,8 @@ def cwt(x, wavelet='gmw', scales='log-piecewise', fs=None, t=None, nv=32,
 
         l1_norm: bool (default True)
             Whether to L1-normalize the CWT, which yields a more representative
-            distribution of energies and component amplitudes than L2 (see [3]).
-            If False (default True), uses L2 norm.
+            distribution of energies and component amplitudes than L2 (see [3],
+            [6]). If False (default True), uses L2 norm.
 
         derivative: bool (default False)
             Whether to compute and return `dWx`. Requires `fs` or `t`.
@@ -138,9 +138,8 @@ def cwt(x, wavelet='gmw', scales='log-piecewise', fs=None, t=None, nv=32,
         E. Brevdo, N.-S. Fučkar, and H.-T. Wu.
         https://arxiv.org/abs/1105.0010
 
-        3. Rectification of the Bias in the Wavelet Power Spectrum.
-        Y. Liu, X. S. Liang, R. H. Weisberg.
-        http://ocg6.marine.usf.edu/~liu/Papers/Liu_etal_2007_JAOT_wavelet.pdf
+        3. How to validate a wavelet filterbank (CWT)? John Muradeli.
+        https://dsp.stackexchange.com/a/86069/50076
 
         4. The Exponential Accuracy of Fourier and Chebyshev Differencing Methods.
         E. Tadmor.
@@ -149,6 +148,10 @@ def cwt(x, wavelet='gmw', scales='log-piecewise', fs=None, t=None, nv=32,
         5. Synchrosqueezing Toolbox, (C) 2014--present. E. Brevdo, G. Thakur.
         https://github.com/ebrevdo/synchrosqueezing/blob/master/synchrosqueezing/
         cwt_fw.m
+
+        6. Rectification of the Bias in the Wavelet Power Spectrum.
+        Y. Liu, X. S. Liang, R. H. Weisberg.
+        http://ocg6.marine.usf.edu/~liu/Papers/Liu_etal_2007_JAOT_wavelet.pdf
     """
     def _vectorized(xh, scales, wavelet, derivative, cache_wavelet):
         if cache_wavelet:
@@ -307,7 +310,7 @@ def _process_args(x, scales, nv, fs, t, nan_checks, wavelet, cache_wavelet):
 
 
 def icwt(Wx, wavelet='gmw', scales='log-piecewise', nv=None, one_int=True,
-         x_len=None, x_mean=0, padtype='zero', rpadded=False, l1_norm=True):
+         x_len=None, x_mean=0, padtype='reflect', rpadded=False, l1_norm=True):
     """The inverse Continuous Wavelet Transform of `Wx`, via double or
     single integral.
 
@@ -332,8 +335,8 @@ def icwt(Wx, wavelet='gmw', scales='log-piecewise', nv=None, one_int=True,
         one_int: bool (default True)
             Whether to use one-integral iCWT or double.
             Current one-integral implementation performs best.
-                - True: Eq 2.6, modified, of [3]. Explained in [4].
-                - False: Eq 4.67 of [1]. Explained in [5].
+                - True: Eq 2.6, modified, of [6]. Explained in [1].
+                - False: Eq 4.67 of [3]. Explained in [2].
 
         x_len: int / None. Length of `x` used in forward CWT, if different
             from Wx.shape[1] (default if None).
@@ -356,25 +359,28 @@ def icwt(Wx, wavelet='gmw', scales='log-piecewise', nv=None, one_int=True,
             The signal, as reconstructed from Wx.
 
     # References:
-        1. Wavelet Tour of Signal Processing, 3rd ed. S. Mallat.
+        1. One integral inverse CWT. John Muradeli.
+        https://dsp.stackexchange.com/a/76239/50076
+
+        2. Inverse CWT derivation. John Muradeli.
+        https://dsp.stackexchange.com/a/71148/50076
+
+        3. Wavelet Tour of Signal Processing, 3rd ed. S. Mallat.
         https://www.di.ens.fr/~mallat/papiers/WaveletTourChap1-2-3.pdf
 
-        2. The Synchrosqueezing algorithm for time-varying spectral analysis:
+        4. Why iCWT may be inexact. John Muradeli.
+        https://dsp.stackexchange.com/a/87104/50076
+
+        5. The Synchrosqueezing algorithm for time-varying spectral analysis:
         robustness properties and new paleoclimate applications. G. Thakur,
         E. Brevdo, N.-S. Fučkar, and H.-T. Wu.
         https://arxiv.org/abs/1105.0010
 
-        3. Synchrosqueezed Wavelet Transforms: a Tool for Empirical Mode
+        6. Synchrosqueezed Wavelet Transforms: a Tool for Empirical Mode
         Decomposition. I. Daubechies, J. Lu, H.T. Wu.
         https://arxiv.org/pdf/0912.2437.pdf
 
-        4. One integral inverse CWT. OverLordGoldDragon.
-        https://dsp.stackexchange.com/a/71274/50076
-
-        5. Inverse CWT derivation. OverLordGoldDragon.
-        https://dsp.stackexchange.com/a/71148/50076
-
-        6. Synchrosqueezing Toolbox, (C) 2014--present. E. Brevdo, G. Thakur.
+        7. Synchrosqueezing Toolbox, (C) 2014--present. E. Brevdo, G. Thakur.
         https://github.com/ebrevdo/synchrosqueezing/blob/master/synchrosqueezing/
         synsq_cwt_fw.m
     """

ssqueezepy/ridge_extraction.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-"""Authors: David Bondesson, OverLordGoldDragon
+"""Authors: David Bondesson, John Muradeli
 
 Ridge extraction from time-frequency representations (STFT, CWT, synchrosqueezed).
 """

ssqueezepy/ssqueezing.py

@@ -184,7 +184,12 @@ def _process_args(Wx, w, fs, t, transform, squeezing, scales, maprange,
                              "tuple with `maprange = 'maximal'` "
                              "(got %s)" % str(maprange))
         ssq_freqs = _compute_associated_frequencies(
-            scales, N, wavelet, ssq_scaletype, maprange, was_padded, dt, transform)
+            scales, N, wavelet, ssq_scaletype, maprange, was_padded, dt,
+            transform)
+    elif transform == 'stft':
+        # removes warning per issue with `infer_scaletype`
+        # future TODO: shouldn't need this
+        ssq_scaletype = 'linear'
     else:
         ssq_scaletype, _ = infer_scaletype(ssq_freqs)