[ENH] Add option to store and return TFR taper weights (#12910)

Co-authored-by: Daniel McCloy <dan@mccloy.info>
Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

Author: 3 people

doc/changes/devel/12910.newfeature.rst

@@ -0,0 +1 @@
+Added the option to return taper weights from :func:`mne.time_frequency.tfr_array_multitaper`, and taper weights are now stored in the :class:`mne.time_frequency.BaseTFR` objects, by `Thomas Binns`_.

mne/time_frequency/multitaper.py

@@ -471,6 +471,7 @@ def tfr_array_multitaper(
     output="complex",
     n_jobs=None,
     *,
+    return_weights=False,
     verbose=None,
 ):
     """Compute Time-Frequency Representation (TFR) using DPSS tapers.
@@ -504,6 +505,11 @@ def tfr_array_multitaper(
           coherence across trials.
     %(n_jobs)s
         The parallelization is implemented across channels.
+    return_weights : bool, default False
+        If True, return the taper weights. Only applies if ``output='complex'`` or
+        ``'phase'``.
+
+        .. versionadded:: 1.10.0
     %(verbose)s
 
     Returns
@@ -520,6 +526,9 @@ def tfr_array_multitaper(
         If ``output`` is ``'avg_power_itc'``, the real values in ``out``
         contain the average power and the imaginary values contain the
         inter-trial coherence: :math:`out = power_{avg} + i * ITC`.
+    weights : array of shape (n_tapers, n_freqs)
+        The taper weights. Only returned if ``output='complex'`` or ``'phase'`` and
+        ``return_weights=True``.
 
     See Also
     --------
@@ -550,6 +559,7 @@ def tfr_array_multitaper(
         use_fft=use_fft,
         decim=decim,
         output=output,
+        return_weights=return_weights,
         n_jobs=n_jobs,
         verbose=verbose,
     )

mne/time_frequency/tests/test_tfr.py

@@ -432,17 +432,21 @@ def test_tfr_morlet():
 def test_dpsswavelet():
     """Test DPSS tapers."""
     freqs = np.arange(5, 25, 3)
-    Ws = _make_dpss(
-        1000, freqs=freqs, n_cycles=freqs / 2.0, time_bandwidth=4.0, zero_mean=True
+    Ws, weights = _make_dpss(
+        1000,
+        freqs=freqs,
+        n_cycles=freqs / 2.0,
+        time_bandwidth=4.0,
+        zero_mean=True,
+        return_weights=True,
     )
 
-    assert len(Ws) == 3  # 3 tapers expected
+    assert np.shape(Ws)[:2] == (3, len(freqs))  # 3 tapers expected
+    assert np.shape(Ws)[:2] == np.shape(weights)  # weights of shape (tapers, freqs)
 
     # Check that zero mean is true
     assert np.abs(np.mean(np.real(Ws[0][0]))) < 1e-5
 
-    assert len(Ws[0]) == len(freqs)  # As many wavelets as asked for
-
 
 @pytest.mark.slowtest
 def test_tfr_multitaper():
@@ -664,6 +668,17 @@ def test_tfr_io(inst, average_tfr, request, tmp_path):
     with tfr.info._unlock():
         tfr.info["meas_date"] = want
     assert tfr_loaded == tfr
+    # test with taper dimension and weights
+    n_tapers = 3  # anything >= 1 should do
+    weights = np.ones((n_tapers, tfr.shape[2]))  # tapers x freqs
+    state = tfr.__getstate__()
+    state["data"] = np.repeat(np.expand_dims(tfr.data, 2), n_tapers, axis=2)  # add dim
+    state["weights"] = weights  # add weights
+    state["dims"] = ("epoch", "channel", "taper", "freq", "time")  # update dims
+    tfr = EpochsTFR(inst=state)
+    tfr.save(fname, overwrite=True)
+    tfr_loaded = read_tfrs(fname)
+    assert tfr_loaded == tfr
     # test overwrite
     with pytest.raises(OSError, match="Destination file exists."):
         tfr.save(fname, overwrite=False)
@@ -722,17 +737,31 @@ def test_average_tfr_init(full_evoked):
         AverageTFR(inst=full_evoked, method="stockwell", freqs=freqs_linspace)
 
 
-def test_epochstfr_init_errors(epochs_tfr):
-    """Test __init__ for EpochsTFR."""
-    state = epochs_tfr.__getstate__()
-    with pytest.raises(ValueError, match="EpochsTFR data should be 4D, got 3"):
-        EpochsTFR(inst=state | dict(data=epochs_tfr.data[..., 0]))
+@pytest.mark.parametrize("inst", ("raw_tfr", "epochs_tfr", "average_tfr"))
+def test_tfr_init_errors(inst, request, average_tfr):
+    """Test __init__ for {Raw,Epochs,Average}TFR."""
+    # Load data
+    inst = _get_inst(inst, request, average_tfr=average_tfr)
+    state = inst.__getstate__()
+    # Prepare for TFRArray object instantiation
+    inst_name = inst.__class__.__name__
+    class_mapping = dict(RawTFR=RawTFR, EpochsTFR=EpochsTFR, AverageTFR=AverageTFR)
+    ndims_mapping = dict(
+        RawTFR=("3D or 4D"), EpochsTFR=("4D or 5D"), AverageTFR=("3D or 4D")
+    )
+    TFR = class_mapping[inst_name]
+    allowed_ndims = ndims_mapping[inst_name]
+    # Check errors caught
+    with pytest.raises(ValueError, match=f".*TFR data should be {allowed_ndims}"):
+        TFR(inst=state | dict(data=inst.data[..., 0]))
+    with pytest.raises(ValueError, match=f".*TFR data should be {allowed_ndims}"):
+        TFR(inst=state | dict(data=np.expand_dims(inst.data, axis=(0, 1))))
     with pytest.raises(ValueError, match="Channel axis of data .* doesn't match info"):
-        EpochsTFR(inst=state | dict(data=epochs_tfr.data[:, :-1]))
+        TFR(inst=state | dict(data=inst.data[..., :-1, :, :]))
     with pytest.raises(ValueError, match="Time axis of data.*doesn't match times attr"):
-        EpochsTFR(inst=state | dict(times=epochs_tfr.times[:-1]))
+        TFR(inst=state | dict(times=inst.times[:-1]))
     with pytest.raises(ValueError, match="Frequency axis of.*doesn't match freqs attr"):
-        EpochsTFR(inst=state | dict(freqs=epochs_tfr.freqs[:-1]))
+        TFR(inst=state | dict(freqs=inst.freqs[:-1]))
 
 
 @pytest.mark.parametrize(
@@ -830,6 +859,25 @@ def test_plot():
     plt.close("all")
 
 
+@pytest.mark.parametrize("output", ("complex", "phase"))
+def test_plot_multitaper_complex_phase(output):
+    """Test TFR plotting of data with a taper dimension."""
+    # Create example data with a taper dimension
+    n_chans, n_tapers, n_freqs, n_times = (3, 4, 2, 3)
+    data = np.random.rand(n_chans, n_tapers, n_freqs, n_times)
+    if output == "complex":
+        data = data + np.random.rand(*data.shape) * 1j  # add imaginary data
+    times = np.arange(n_times)
+    freqs = np.arange(n_freqs)
+    weights = np.random.rand(n_tapers, n_freqs)
+    info = mne.create_info(n_chans, 1000.0, "eeg")
+    tfr = AverageTFRArray(
+        info=info, data=data, times=times, freqs=freqs, weights=weights
+    )
+    # Check that plotting works
+    tfr.plot()
+
+
 @pytest.mark.parametrize(
     "timefreqs,title,combine",
     (
@@ -1154,6 +1202,15 @@ def test_averaging_epochsTFR():
     ):
         power.average(method=np.mean)
 
+    # Check it doesn't run for taper spectra
+    tapered = epochs.compute_tfr(
+        method="multitaper", freqs=freqs, n_cycles=n_cycles, output="complex"
+    )
+    with pytest.raises(
+        NotImplementedError, match=r"Averaging multitaper tapers .* is not supported."
+    ):
+        tapered.average()
+
 
 def test_averaging_freqsandtimes_epochsTFR():
     """Test that EpochsTFR averaging freqs methods work."""
@@ -1258,12 +1315,15 @@ def test_to_data_frame():
     ch_names = ["EEG 001", "EEG 002", "EEG 003", "EEG 004"]
     n_picks = len(ch_names)
     ch_types = ["eeg"] * n_picks
+    n_tapers = 2
     n_freqs = 5
     n_times = 6
-    data = np.random.rand(n_epos, n_picks, n_freqs, n_times)
-    times = np.arange(6)
+    data = np.random.rand(n_epos, n_picks, n_tapers, n_freqs, n_times)
+    times = np.arange(n_times)
     srate = 1000.0
-    freqs = np.arange(5)
+    freqs = np.arange(n_freqs)
+    tapers = np.arange(n_tapers)
+    weights = np.ones((n_tapers, n_freqs))
     events = np.zeros((n_epos, 3), dtype=int)
     events[:, 0] = np.arange(n_epos)
     events[:, 2] = np.arange(5, 5 + n_epos)
@@ -1276,6 +1336,7 @@ def test_to_data_frame():
         freqs=freqs,
         events=events,
         event_id=event_id,
+        weights=weights,
     )
     # test index checking
     with pytest.raises(ValueError, match="options. Valid index options are"):
@@ -1287,32 +1348,51 @@ def test_to_data_frame():
     # test wide format
     df_wide = tfr.to_data_frame()
     assert all(np.isin(tfr.ch_names, df_wide.columns))
-    assert all(np.isin(["time", "condition", "freq", "epoch"], df_wide.columns))
+    assert all(
+        np.isin(["time", "condition", "freq", "epoch", "taper"], df_wide.columns)
+    )
     # test long format
     df_long = tfr.to_data_frame(long_format=True)
-    expected = ("condition", "epoch", "freq", "time", "channel", "ch_type", "value")
+    expected = (
+        "condition",
+        "epoch",
+        "freq",
+        "time",
+        "channel",
+        "ch_type",
+        "value",
+        "taper",
+    )
     assert set(expected) == set(df_long.columns)
     assert set(tfr.ch_names) == set(df_long["channel"])
     assert len(df_long) == tfr.data.size
     # test long format w/ index
     df_long = tfr.to_data_frame(long_format=True, index=["freq"])
     del df_wide, df_long
     # test whether data is in correct shape
-    df = tfr.to_data_frame(index=["condition", "epoch", "freq", "time"])
+    df = tfr.to_data_frame(index=["condition", "epoch", "taper", "freq", "time"])
     data = tfr.data
     assert_array_equal(df.values[:, 0], data[:, 0, :, :].reshape(1, -1).squeeze())
     # compare arbitrary observation:
     assert (
-        df.loc[("he", slice(None), freqs[1], times[2]), ch_names[3]].iat[0]
-        == data[1, 3, 1, 2]
+        df.loc[("he", slice(None), tapers[1], freqs[1], times[2]), ch_names[3]].iat[0]
+        == data[1, 3, 1, 1, 2]
     )
 
     # Check also for AverageTFR:
+    # (remove taper dimension before averaging)
+    state = tfr.__getstate__()
+    state["data"] = state["data"][:, :, 0]
+    state["dims"] = ("epoch", "channel", "freq", "time")
+    state["weights"] = None
+    tfr = EpochsTFR(inst=state)
     tfr = tfr.average()
     with pytest.raises(ValueError, match="options. Valid index options are"):
         tfr.to_data_frame(index=["epoch", "condition"])
     with pytest.raises(ValueError, match='"epoch" is not a valid option'):
         tfr.to_data_frame(index="epoch")
+    with pytest.raises(ValueError, match='"taper" is not a valid option'):
+        tfr.to_data_frame(index="taper")
     with pytest.raises(TypeError, match="index must be `None` or a string "):
         tfr.to_data_frame(index=np.arange(400))
     # test wide format
@@ -1348,11 +1428,13 @@ def test_to_data_frame_index(index):
     ch_names = ["EEG 001", "EEG 002", "EEG 003", "EEG 004"]
     n_picks = len(ch_names)
     ch_types = ["eeg"] * n_picks
+    n_tapers = 2
     n_freqs = 5
     n_times = 6
-    data = np.random.rand(n_epos, n_picks, n_freqs, n_times)
-    times = np.arange(6)
-    freqs = np.arange(5)
+    data = np.random.rand(n_epos, n_picks, n_tapers, n_freqs, n_times)
+    times = np.arange(n_times)
+    freqs = np.arange(n_freqs)
+    weights = np.ones((n_tapers, n_freqs))
     events = np.zeros((n_epos, 3), dtype=int)
     events[:, 0] = np.arange(n_epos)
     events[:, 2] = np.arange(5, 8)
@@ -1365,14 +1447,15 @@ def test_to_data_frame_index(index):
         freqs=freqs,
         events=events,
         event_id=event_id,
+        weights=weights,
     )
     df = tfr.to_data_frame(picks=[0, 2, 3], index=index)
     # test index order/hierarchy preservation
     if not isinstance(index, list):
         index = [index]
     assert list(df.index.names) == index
     # test that non-indexed data were present as columns
-    non_index = list(set(["condition", "time", "freq", "epoch"]) - set(index))
+    non_index = list(set(["condition", "time", "freq", "taper", "epoch"]) - set(index))
     if len(non_index):
         assert all(np.isin(non_index, df.columns))
 
@@ -1538,7 +1621,8 @@ def test_epochs_compute_tfr_stockwell(epochs, freqs, return_itc):
 def test_epochs_compute_tfr_multitaper_complex_phase(epochs, output):
     """Test Epochs.compute_tfr(output="complex"/"phase")."""
     tfr = epochs.compute_tfr("multitaper", freqs_linspace, output=output)
-    assert len(tfr.shape) == 5
+    assert len(tfr.shape) == 5  # epoch x channel x taper x freq x time
+    assert tfr.weights.shape == tfr.shape[2:4]  # check weights and coeffs shapes match
 
 
 @pytest.mark.parametrize("copy", (False, True))
@@ -1550,6 +1634,42 @@ def test_epochstfr_iter_evoked(epochs_tfr, copy):
     assert avgs[0].comment == str(epochs_tfr.events[0, -1])
 
 
+@pytest.mark.parametrize("obj_type", ("raw", "epochs", "evoked"))
+def test_tfrarray_tapered_spectra(obj_type):
+    """Test {Raw,Epochs,Average}TFRArray instantiation with tapered spectra."""
+    # Create example data with a taper dimension
+    n_epochs, n_chans, n_tapers, n_freqs, n_times = (5, 3, 4, 2, 6)
+    data_shape = (n_chans, n_tapers, n_freqs, n_times)
+    if obj_type == "epochs":
+        data_shape = (n_epochs,) + data_shape
+    data = np.random.rand(*data_shape)
+    times = np.arange(n_times)
+    freqs = np.arange(n_freqs)
+    weights = np.random.rand(n_tapers, n_freqs)
+    info = mne.create_info(n_chans, 1000.0, "eeg")
+    # Prepare for TFRArray object instantiation
+    defaults = dict(info=info, data=data, times=times, freqs=freqs)
+    class_mapping = dict(raw=RawTFRArray, epochs=EpochsTFRArray, evoked=AverageTFRArray)
+    TFRArray = class_mapping[obj_type]
+    # Check TFRArray instantiation runs with good data
+    TFRArray(**defaults, weights=weights)
+    # Check taper dimension but no weights caught
+    with pytest.raises(
+        ValueError, match="Taper dimension in data, but no weights found."
+    ):
+        TFRArray(**defaults)
+    # Check mismatching n_taper in weights caught
+    with pytest.raises(
+        ValueError, match=r"Taper axis .* doesn't match weights attribute"
+    ):
+        TFRArray(**defaults, weights=weights[:-1])
+    # Check mismatching n_freq in weights caught
+    with pytest.raises(
+        ValueError, match=r"Frequency axis .* doesn't match weights attribute"
+    ):
+        TFRArray(**defaults, weights=weights[:, :-1])
+
+
 def test_tfr_proj(epochs):
     """Test `compute_tfr(proj=True)`."""
     epochs.compute_tfr(method="morlet", freqs=freqs_linspace, proj=True)
@@ -1731,3 +1851,52 @@ def test_tfr_plot_topomap(inst, ch_type, full_average_tfr, request):
     assert re.match(
         rf"Average over \d{{1,3}} {ch_type} channels\.", popup_fig.axes[0].get_title()
     )
+
+
+@pytest.mark.parametrize("output", ("complex", "phase"))
+def test_tfr_topo_plotting_multitaper_complex_phase(output, evoked):
+    """Test plot_joint/topo/topomap() for data with a taper dimension."""
+    # Compute TFR with taper dimension
+    tfr = evoked.compute_tfr(
+        method="multitaper", freqs=freqs_linspace, n_cycles=4, output=output
+    )
+    # Check that plotting works
+    tfr.plot_joint(topomap_args=dict(res=8, contours=0, sensors=False))  # for speed
+    tfr.plot_topo()
+    tfr.plot_topomap()
+
+
+def test_combine_tfr_error_catch(average_tfr):
+    """Test combine_tfr() catches errors."""
+    # check unrecognised weights string caught
+    with pytest.raises(ValueError, match='Weights must be .* "nave" or "equal"'):
+        combine_tfr([average_tfr, average_tfr], weights="foo")
+    # check bad weights size caught
+    with pytest.raises(ValueError, match="Weights must be the same size as all_tfr"):
+        combine_tfr([average_tfr, average_tfr], weights=[1, 1, 1])
+    # check different channel names caught
+    state = average_tfr.__getstate__()
+    new_info = average_tfr.info.copy()
+    average_tfr_bad = AverageTFR(
+        inst=state | dict(info=new_info.rename_channels({new_info.ch_names[0]: "foo"}))
+    )
+    with pytest.raises(AssertionError, match=".* do not contain the same channels"):
+        combine_tfr([average_tfr, average_tfr_bad])
+    # check different times caught
+    average_tfr_bad = AverageTFR(inst=state | dict(times=average_tfr.times + 1))
+    with pytest.raises(
+        AssertionError, match=".* do not contain the same time instants"
+    ):
+        combine_tfr([average_tfr, average_tfr_bad])
+    # check taper dim caught
+    n_tapers = 3  # anything >= 1 should do
+    weights = np.ones((n_tapers, average_tfr.shape[1]))  # tapers x freqs
+    state["data"] = np.repeat(np.expand_dims(average_tfr.data, 1), n_tapers, axis=1)
+    state["weights"] = weights
+    state["dims"] = ("channel", "taper", "freq", "time")
+    average_tfr_taper = AverageTFR(inst=state)
+    with pytest.raises(
+        NotImplementedError,
+        match="Aggregating multitaper tapers across TFR datasets is not supported.",
+    ):
+        combine_tfr([average_tfr_taper, average_tfr_taper])