[ENH] Tiny changes to dss.dss0() (#71)

citation.cff

@@ -5,7 +5,7 @@ authors:
   given-names: "Nicolas"
   orcid: "https://orcid.org/0000-0003-1495-561X"
 title: "MEEGkit"
-version: 0.1.4
-doi: 10.5281/zenodo.5643659
-date-released: 2021-10-15
+version: 0.1.5
+doi: 10.5281/zenodo.10210992
+date-released: 2023-11-27
 url: "https://github.com/nbara/python-meegkit"

meegkit/__init__.py

@@ -1,5 +1,5 @@
 """M/EEG denoising utilities in python."""
-__version__ = "0.1.4"
+__version__ = "0.1.5"
 
 from . import asr, cca, detrend, dss, lof, ress, sns, star, trca, tspca, utils
 

meegkit/asr.py

@@ -14,7 +14,7 @@
     pyriemann = None
 
 
-class ASR():
+class ASR:
     """Artifact Subspace Reconstruction.
 
     Artifact subspace reconstruction (ASR) is an automatic, online,

meegkit/cca.py

@@ -63,12 +63,12 @@ def mcca(C, n_channels, n_keep=[]):
         W = whiten_nt(CC, keep=True)
         A[ix0:ix1, ix0:ix1] = W
 
-    C = A.T.dot(C.dot(A))
+    C = A.T @ C @ A
 
     # final PCA
     V, d = pca(C, thresh=None)  # don't threshold the PCA to keep n_channels
-    A = A.dot(V)
-    C = V.T.dot(C.dot(V))
+    A = A @ V
+    C = V.T @ C @ V
     scores = np.diag(C)
 
     AA = []

meegkit/detrend.py

@@ -159,7 +159,7 @@ def regress(x, r, w=None, threshold=1e-7, return_mean=False):
         yy = demean(x)
 
         # PCA
-        V, _ = pca(rr.T.dot(rr), thresh=threshold)
+        V, _ = pca(rr.T @ rr, thresh=threshold)
         rrr = rr.dot(V)
 
         # Regression (OLS)

meegkit/dss.py

@@ -68,7 +68,7 @@ def dss1(X, weights=None, keep1=None, keep2=1e-12):
     return todss, fromdss, pwr0, pwr1
 
 
-def dss0(c0, c1, keep1=None, keep2=1e-9):
+def dss0(c0, c1, keep1=None, keep2=1e-9, return_unmixing=True):
     """DSS base function.
 
     This function allows specifying arbitrary bias functions (as compared to
@@ -84,13 +84,16 @@ def dss0(c0, c1, keep1=None, keep2=1e-9):
         Number of PCs to retain (default=None, which keeps all).
     keep2: float
         Ignore PCs smaller than keep2 (default=1e-9).
+    return_unmixing : bool
+        If True (default), return the unmixing matrix.
 
     Returns
     -------
-    todss: array, shape=(n_dss_components, n_chans)
+    todss: array, shape=(n_chans, n_dss_components)
         Matrix to convert X to normalized DSS components.
-    fromdss : array, shape=()
-        Matrix to transform back to original space.
+    fromdss : array, shape=(n_dss_components, n_chans)
+        Matrix to transform back to original space. Only returned if
+        ``return_unmixing`` is True.
     pwr0: array
         Power per component (baseline).
     pwr1: array
@@ -101,46 +104,46 @@ def dss0(c0, c1, keep1=None, keep2=1e-9):
     The data mean is NOT removed prior to processing.
 
     """
-    if c0 is None or c1 is None:
-        raise AttributeError("dss0 needs at least two arguments")
-    if c0.shape != c1.shape:
-        raise AttributeError("c0 and c1 should have same size")
-    if c0.shape[0] != c0.shape[1]:
-        raise AttributeError("c0 should be square")
-    if np.any(np.isnan(c0)) or np.any(np.isinf(c0)):
-        raise ValueError("NaN or INF in c0")
-    if np.any(np.isnan(c1)) or np.any(np.isinf(c1)):
-        raise ValueError("NaN or INF in c1")
+    # Check size and squareness
+    assert c0.shape == c1.shape == (c0.shape[0], c0.shape[0]), \
+        "c0 and c1 should have the same size, and be square"
+
+    # Check for NaN or INF
+    assert not (np.any(np.isnan(c0)) or np.any(np.isinf(c0))), "NaN or INF in c0"
+    assert not (np.any(np.isnan(c1)) or np.any(np.isinf(c1))), "NaN or INF in c1"
 
     # derive PCA and whitening matrix from unbiased covariance
     eigvec0, eigval0 = pca(c0, max_comps=keep1, thresh=keep2)
 
     # apply whitening and PCA matrices to the biased covariance
     # (== covariance of bias whitened data)
-    W = np.sqrt(1. / eigval0)  # diagonal of whitening matrix
+    W = np.diag(np.sqrt(1. / eigval0))  # diagonal of whitening matrix
 
     # c1 is projected into whitened PCA space of data channels
-    c2 = (W * eigvec0).T.dot(c1).dot(eigvec0) * W
+    c2 = (eigvec0 @ W).T @ c1 @ (eigvec0 @ W)
 
     # proj. matrix from whitened data space to a space maximizing bias
     eigvec2, eigval2 = pca(c2, max_comps=keep1, thresh=keep2)
 
     # DSS matrix (raw data to normalized DSS)
-    todss = (W[np.newaxis, :] * eigvec0).dot(eigvec2)
-    fromdss = linalg.pinv(todss)
+    todss = eigvec0 @ W @ eigvec2
 
     # Normalise DSS matrix
-    N = np.sqrt(1. / np.diag(np.dot(np.dot(todss.T, c0), todss)))
-    todss = todss * N
+    N = np.sqrt(np.diag(todss.T @ c0 @ todss))
+    todss /= N
 
-    pwr0 = np.sqrt(np.sum(np.dot(c0, todss) ** 2, axis=0))
-    pwr1 = np.sqrt(np.sum(np.dot(c1, todss) ** 2, axis=0))
+    pwr0 = np.sqrt(np.sum((c0 @ todss) ** 2, axis=0))
+    pwr1 = np.sqrt(np.sum((c1 @ todss) ** 2, axis=0))
 
     # Return data
     # next line equiv. to: np.array([np.dot(todss, ep) for ep in data])
     # dss_data = np.einsum('ij,hjk->hik', todss, data)
 
-    return todss, fromdss, pwr0, pwr1
+    if return_unmixing:
+        fromdss = linalg.pinv(todss)
+        return todss, fromdss, pwr0, pwr1
+    else:
+        return todss, pwr0, pwr1
 
 
 def dss_line(X, fline, sfreq, nremove=1, nfft=1024, nkeep=None, blocksize=None,
@@ -373,7 +376,7 @@ def nan_basic_interp(array):
             ax.flat[3].plot(np.arange(iterations + 1), aggr_resid, marker="o")
             ax.flat[3].set_title("Iterations")
 
-            f.set_tight_layout(True)
+            plt.tight_layout()
             plt.savefig(f"{prefix}_{iterations:03}.png")
             plt.close("all")
 

meegkit/lof.py

@@ -7,7 +7,7 @@
 from sklearn.neighbors import LocalOutlierFactor
 
 
-class LOF():
+class LOF:
     """Local Outlier Factor.
 
     Local Outlier Factor (LOF) is an automatic, density-based outlier detection

meegkit/trca.py

@@ -134,8 +134,6 @@ def predict(self, X):
         ----------
         X: array, shape=(n_samples, n_chans[, n_trials])
             Test data.
-        model: dict
-            Fitted model to be used in testing phase.
 
         Returns
         -------

meegkit/tspca.py

@@ -192,8 +192,8 @@ def tsr(X, R, shifts=None, wX=None, wR=None, keep=None, thresh=1e-12):
     # TSPCA: clean x by removing regression on time-shifted refs
     y = np.zeros((n_samples_X, n_chans_X, n_trials_X))
     for t in np.arange(n_trials_X):
-        r = multishift(R[..., t], shifts, reshape=True)
-        y[..., t] = X[:z.shape[0], :, t] - (r @ regression)
+        z = multishift(R[..., t], shifts, reshape=True) @ regression
+        y[..., t] = X[:z.shape[0], :, t] - z
 
     y, mean2 = demean(y, wX, return_mean=True, inplace=True)
     idx = np.arange(offset1, initial_samples - offset2)

meegkit/utils/auditory.py

@@ -56,7 +56,7 @@ def erbspace(flow, fhigh, n):
     return y, bw
 
 
-class GammatoneFilterbank():
+class GammatoneFilterbank:
     """Gammatone Filterbank.
 
     This class computes the filter coefficients for a bank of Gammatone

meegkit/utils/covariances.py

@@ -372,8 +372,7 @@ def pca(cov, max_comps=None, thresh=0):
 
     var = 100 * d.sum() / p0
     if var < 99:
-        print("[PCA] Explained variance of selected components : {:.2f}%".
-              format(var))
+        print(f"[PCA] Explained variance of selected components : {var:.2f}%")
 
     return V, d
 

meegkit/utils/denoise.py

@@ -162,7 +162,7 @@ def find_outlier_trials(X, thresh=None, show=True):
     thresh : float or array of floats
         Keep trials less than thresh from mean.
     show : bool
-        If true plot trial deviations before and after.
+        If True (default), plot trial deviations before and after.
 
     Returns
     -------
@@ -173,9 +173,11 @@ def find_outlier_trials(X, thresh=None, show=True):
 
     """
     if thresh is None:
-        thresh = [np.inf]
-    elif isinstance(thresh, float) or isinstance(thresh, int):
-        thresh = [thresh]
+        thresh = np.array([np.inf])
+    elif isinstance(thresh, (float, int)):
+        thresh = np.array([thresh])
+    else:
+        thresh = np.asarray(thresh)
 
     if X.ndim > 3:
         raise ValueError("X should be 2D or 3D")
@@ -206,7 +208,7 @@ def find_outlier_trials(X, thresh=None, show=True):
         ax1.axhline(y=thresh[0], color="grey", linestyle=":")
         ax1.set_xlabel("Trial #")
         ax1.set_ylabel("Normalized deviation from mean")
-        ax1.set_title("Before, " + str(len(d)), fontsize=10)
+        ax1.set_title("Before, " + str(len(d)))
         ax1.set_xlim(0, len(d) + 1)
         plt.draw()
 
@@ -215,19 +217,18 @@ def find_outlier_trials(X, thresh=None, show=True):
         _, dd = find_outlier_trials(X[:, idx], None, False)
         ax2.plot(dd, ls="-")
         ax2.set_xlabel("Trial #")
-        ax2.set_title("After, " + str(len(idx)), fontsize=10)
+        ax2.set_title("After, " + str(len(idx)))
         ax2.yaxis.tick_right()
         ax2.set_xlim(0, len(idx) + 1)
         plt.show()
 
-    thresh.pop(0)
+    thresh = thresh[1:]
     if thresh:
-        bads2, _ = find_outlier_trials(X[:, idx], thresh, show)
-        idx2 = idx[bads2]
-        idx = np.setdiff1d(idx, idx2)
+        bads, _ = find_outlier_trials(X[:, idx], thresh, show)
+        idx = np.setdiff1d(idx, idx[bads])
 
-    bads = []
+    bads_accumulated = []
     if len(idx) < n_trials:
-        bads = np.setdiff1d(range(n_trials), idx)
+        bads_accumulated = np.setdiff1d(range(n_trials), idx)
 
-    return bads, d
+    return bads_accumulated, d

meegkit/utils/sig.py

@@ -365,9 +365,7 @@ def gaussfilt(data, srate, f, fwhm, n_harm=1, shift=0, return_empvals=False, sho
         plt.plot(hz, fx, "o-")
         plt.xlim([0, None])
 
-        title = "Requested: {}, {} Hz\nEmpirical: {}, {} Hz".format(
-            f, fwhm, empVals[0], empVals[1]
-        )
+        title = f"Requested: {f}, {fwhm} Hz\nEmpirical: {empVals[0]}, {empVals[1]} Hz"
         plt.title(title)
         plt.xlabel("Frequency (Hz)")
         plt.ylabel("Amplitude gain")
@@ -515,8 +513,7 @@ def stmcb(x, u_in=None, q=None, p=None, niter=5, a_in=None):
     else:
         if len(u_in) != len(x):
             raise ValueError(
-                "stmcb: u_in and x must be of the same size: {} != {}".format(
-                    len(u_in), len(x)))
+                f"stmcb: u_in and x must be of the same size: {len(u_in)} != {len(x)}")
         if a_in is None:
             q = 0
             _, a_in = prony(x, q, p)

tests/test_ress.py

@@ -66,7 +66,8 @@ def test_ress(target, n_trials, peak_width, neig_width, neig_freq, show=False):
                                n_chans=n_chans, freq=target, sfreq=sfreq,
                                show=False)
     r = ress.RESS(sfreq=sfreq, peak_freq=target, neig_freq=neig_freq,
-        peak_width=peak_width, neig_width=neig_width, n_keep=n_keep, compute_unmixing=True)
+        peak_width=peak_width, neig_width=neig_width, n_keep=n_keep,
+        compute_unmixing=True)
     out = r.fit_transform(data)
 
     nfft = 500

tests/test_tspca.py

@@ -22,14 +22,13 @@ def test_tspca_sns_dss():  # TODO
 
     # remove means
     noisy_data = demean(data)
-    demean(ref)
 
     # Apply TSPCA
     # -------------------------------------------------------------------------
-    # shifts = np.arange(-50, 51)
-    # print('TSPCA...')
-    # y_tspca, idx = tspca.tsr(noisy_data, noisy_ref, shifts)[0:2]
-    # print('\b OK!')
+    shifts = np.arange(-50, 51)
+    print("TSPCA...")
+    y_tspca, idx = tspca.tsr(noisy_data, ref, shifts)[0:2]
+    print("\b OK!")
     y_tspca = noisy_data
 
     # Apply SNS
@@ -84,7 +83,11 @@ def test_tsr(show=True):
         ax[0].plot(x[:500, 0], ":", label="real signal")
         ax[1].plot((y - x)[:500], label="residual")
         ax[0].legend()
-        ax[1].legend()
+        ax[1].set_xlabel("time (samples)")
+        ax[0].set_title("signals")
+        ax[1].set_title("residuals")
+        plt.tight_layout()
+        # ax[1].legend()
         # plt.show()
 
     # Test residual almost 0.0
@@ -103,9 +106,11 @@ def test_tsr(show=True):
         ax[1].plot(x[:500, 0], "grey", label="real signal")
         ax[1].plot(y[:500, 0], ":", label="recovered signal")
         ax[2].plot((signal - y)[:500, 0], label="before - after")
-        ax[0].legend()
+        # ax[0].legend()
         ax[1].legend()
         ax[2].legend()
+        ax[1].set_xlabel("time (samples)")
+        plt.tight_layout()
         plt.show()
 
 if __name__ == "__main__":

tests/test_utils.py

@@ -238,7 +238,7 @@ def test_outliers(show=False):
     idx, _ = find_outlier_trials(x, 2, show=show)
     np.testing.assert_array_equal(idx, np.arange(5))
 
-    idx, _ = find_outlier_trials(x, [2, 2], show=show)
+    idx, _ = find_outlier_trials(x, [2, 2], show=True)
     np.testing.assert_array_equal(idx, np.arange(5))
 
     idx = find_outlier_samples(x, 5)