dclab 0.67.0__cp314-cp314t-macosx_10_13_x86_64.whl

dclab/rtdc_dataset/feat_anc_core/__init__.py

@@ -0,0 +1,24 @@
+from .ancillary_feature import AncillaryFeature  # noqa: F401
+from . import af_basic
+from . import af_emodulus
+from . import af_fl_max_ctc
+from . import af_image_contour
+from . import af_ml_class
+
+
+#: features whose computation is fast
+FEATURES_RAPID = [
+    "area_ratio",
+    "area_um",
+    "aspect",
+    "deform",
+    "index",
+    "time",
+]
+
+
+af_basic.register()
+af_emodulus.register()
+af_fl_max_ctc.register()
+af_image_contour.register()
+af_ml_class.register()

dclab/rtdc_dataset/feat_anc_core/af_basic.py

@@ -0,0 +1,75 @@
+import numpy as np
+
+from .ancillary_feature import AncillaryFeature
+
+
+def compute_area_ratio(mm):
+    valid = mm["area_msd"] != 0
+    out = np.nan * np.ones(len(mm), dtype=float)
+    return np.divide(mm["area_cvx"], mm["area_msd"], where=valid, out=out)
+
+
+def compute_area_um(mm):
+    pxs = mm.config["imaging"]["pixel size"]
+    return mm["area_cvx"] * pxs**2
+
+
+def compute_aspect(mm):
+    """Compute the aspect ratio of the bounding box
+
+    Notes
+    -----
+    If the cell is elongated along the channel, i.e.
+    `size_x` is larger than `size_y`, then the aspect
+    ratio is larger than 1.
+    """
+    out = np.nan * np.ones(len(mm), dtype=float)
+    valid = mm["size_y"] != 0
+    # parallel to flow, perpendicular to flow
+    return np.divide(mm["size_x"], mm["size_y"], where=valid, out=out)
+
+
+def compute_deform(mm):
+    return 1 - mm["circ"]
+
+
+def compute_index(mm):
+    return np.arange(1, len(mm)+1)
+
+
+def compute_time(mm):
+    fr = mm.config["imaging"]["frame rate"]
+    # Since version 0.47.8, we don't "normalize" the time anymore
+    # with the information from mm["frame"][0]. This is important
+    # for cases where it is important to know the time elapsed before
+    # the first event was recorded (issue #207).
+    return np.array(mm["frame"], dtype=float) / fr
+
+
+AncillaryFeature(feature_name="time",
+                 method=compute_time,
+                 req_config=[["imaging", ["frame rate"]]],
+                 req_features=["frame"])
+
+
+AncillaryFeature(feature_name="index",
+                 method=compute_index)
+
+
+def register():
+    AncillaryFeature(feature_name="area_ratio",
+                     method=compute_area_ratio,
+                     req_features=["area_cvx", "area_msd"])
+
+    AncillaryFeature(feature_name="area_um",
+                     method=compute_area_um,
+                     req_config=[["imaging", ["pixel size"]]],
+                     req_features=["area_cvx"])
+
+    AncillaryFeature(feature_name="aspect",
+                     method=compute_aspect,
+                     req_features=["size_x", "size_y"])
+
+    AncillaryFeature(feature_name="deform",
+                     method=compute_deform,
+                     req_features=["circ"])

dclab/rtdc_dataset/feat_anc_core/af_emodulus.py

@@ -0,0 +1,160 @@
+import warnings
+
+from ... import features
+
+from .ancillary_feature import AncillaryFeature
+
+
+def compute_emodulus(mm):
+    """Wrapper function for computing the Young's modulus
+
+    Please take a look at the docs :ref:`sec_emodulus_usage`
+    for more details on the three cases A, B, and C.
+    There are also some sanity checks taking place here.
+    """
+    calccfg = mm.config["calculation"]
+
+    medium = calccfg.get("emodulus medium", "other").lower()
+    temperature = calccfg.get("emodulus temperature", None)
+    viscosity = calccfg.get("emodulus viscosity", None)
+
+    if viscosity is not None and medium == "other":
+        # sanity checks
+        if temperature is not None:
+            warnings.warn("The 'emodulus temperature' configuration key is "
+                          "ignored if the 'emodulus viscosity' key is set!")
+        # Case B from the docs
+        return compute_emodulus_visc_only(mm)
+    else:
+        # sanity checks
+        if not isinstance(medium, str):
+            raise ValueError(
+                f"'emodulus medium' must be a string, got '{medium}'!")
+        if medium not in features.emodulus.viscosity.KNOWN_MEDIA:
+            raise ValueError(
+                f"Only the following media are supported: "
+                f"{features.emodulus.viscosity.KNOWN_MEDIA}, got '{medium}'!")
+        if viscosity is not None:
+            raise ValueError("You must not set the 'emodulus viscosity' "
+                             "configuration keyword for known media!")
+        # warnings
+        if "emodulus viscosity model" not in calccfg:
+            warnings.warn("Please specify the 'emodulus viscosity model' "
+                          "key in the 'calculation' config segion, falling "
+                          "back to 'herold-2017'!",
+                          DeprecationWarning)
+        # actual function calls
+        if temperature is not None:
+            # case C from the docs
+            temperature = mm.config["calculation"]["emodulus temperature"]
+            return compute_emodulus_known_media(mm, temperature=temperature)
+        elif "temp" in mm:
+            # case A from the docs
+            return compute_emodulus_known_media(mm, temperature=mm["temp"])
+
+
+def compute_emodulus_known_media(mm, temperature):
+    """Only use known media and one temperature for all"""
+    calccfg = mm.config["calculation"]
+    # compute elastic modulus
+    emod = features.emodulus.get_emodulus(
+        area_um=mm["area_um"],
+        deform=mm["deform"],
+        medium=calccfg["emodulus medium"],
+        channel_width=mm.config["setup"]["channel width"],
+        flow_rate=mm.config["setup"]["flow rate"],
+        px_um=mm.config["imaging"]["pixel size"],
+        temperature=temperature,
+        lut_data=calccfg["emodulus lut"],
+        visc_model=calccfg.get("emodulus viscosity model", "herold-2017"),
+    )
+    return emod
+
+
+def compute_emodulus_visc_only(mm):
+    """The user entered the viscosity directly"""
+    calccfg = mm.config["calculation"]
+    # compute elastic modulus
+    emod = features.emodulus.get_emodulus(
+        area_um=mm["area_um"],
+        deform=mm["deform"],
+        medium=calccfg["emodulus viscosity"],
+        channel_width=mm.config["setup"]["channel width"],
+        flow_rate=mm.config["setup"]["flow rate"],
+        px_um=mm.config["imaging"]["pixel size"],
+        temperature=None,
+        visc_model=None,
+        lut_data=calccfg["emodulus lut"],
+    )
+    return emod
+
+
+def is_channel(mm):
+    """Check whether the measurement was performed in the channel
+
+    If the chip region is not set, then it is assumed to be a
+    channel measurement (for backwards compatibility and user-
+    friendliness).
+    """
+    if "setup" in mm.config and "chip region" in mm.config["setup"]:
+        region = mm.config["setup"]["chip region"]
+        if region == "channel":
+            # measured in the channel
+            return True
+        else:
+            # measured in the reservoir
+            return False
+    else:
+        # This might be a testing dictionary or someone who is
+        # playing around with data. Avoid disappointments here.
+        return True
+
+
+def register():
+    # Please note that registering these things is a delicate business,
+    # because the priority has to be chosen carefully.
+    # Note that here we have not included the "emodulus viscosity model"
+    # configuration keyword. This is checked in the `compute_emodulus`
+    # method above and a deprecation warning is issued, so old code
+    # does not break immediately.
+    for pr, vm in [(1, ["emodulus viscosity model"]),
+                   (0, [])  # this is deprecated and should be removed!
+                   ]:
+        AncillaryFeature(feature_name="emodulus",
+                         method=compute_emodulus,
+                         data="case C",
+                         req_features=["area_um", "deform"],
+                         req_config=[["calculation", vm + [
+                                        "emodulus lut",
+                                        "emodulus medium",
+                                        "emodulus temperature"]],
+                                     ["imaging", ["pixel size"]],
+                                     ["setup", ["flow rate", "channel width"]]
+                                     ],
+                         req_func=is_channel,
+                         priority=4 + pr)
+        AncillaryFeature(feature_name="emodulus",
+                         data="case A",
+                         method=compute_emodulus,
+                         req_features=["area_um", "deform", "temp"],
+                         req_config=[["calculation", vm + [
+                                        "emodulus lut",
+                                        "emodulus medium"]],
+                                     ["imaging", ["pixel size"]],
+                                     ["setup", ["flow rate", "channel width"]]
+                                     ],
+                         req_func=is_channel,
+                         priority=0 + pr)
+
+    AncillaryFeature(feature_name="emodulus",
+                     data="case B",
+                     method=compute_emodulus,
+                     req_features=["area_um", "deform"],
+                     req_config=[["calculation", vm + [
+                                    "emodulus lut",
+                                    "emodulus viscosity"]],
+                                 ["imaging", ["pixel size"]],
+                                 ["setup", ["flow rate", "channel width"]]
+                                 ],
+                     req_func=is_channel,
+                     priority=2)

dclab/rtdc_dataset/feat_anc_core/af_fl_max_ctc.py

@@ -0,0 +1,133 @@
+
+
+from ... import features
+from .ancillary_feature import AncillaryFeature
+
+
+class MissingCrosstalkMatrixElementsError(BaseException):
+    pass
+
+
+def compute_ctc(mm, fl_channel):
+    if "fl1_max" in mm:
+        fl1 = mm["fl1_max"]
+    else:
+        fl1 = 0
+
+    if "fl2_max" in mm:
+        fl2 = mm["fl2_max"]
+    else:
+        fl2 = 0
+
+    if "fl3_max" in mm:
+        fl3 = mm["fl3_max"]
+    else:
+        fl3 = 0
+
+    ctdict = {}
+
+    for i in [1, 2, 3]:
+        for j in [1, 2, 3]:
+            if i == j:
+                continue
+            key = "crosstalk fl{}{}".format(i, j)
+            par = "ct{}{}".format(i, j)
+            if key in mm.config["calculation"]:
+                ctdict[par] = mm.config["calculation"][key]
+
+    if ("fl1_max" in mm and
+        "fl2_max" in mm and
+        "fl3_max" in mm and
+        ("ct12" not in ctdict or
+         "ct13" not in ctdict or
+         "ct21" not in ctdict or
+         "ct23" not in ctdict or
+         "ct31" not in ctdict or
+         "ct32" not in ctdict)):
+        msg = "{}, has fl1_max, fl2_max, and fl3_max,".format(mm) \
+              + " but not all crosstalk matrix elements are" \
+              + " defined in the 'calculation' configuration section."
+        raise MissingCrosstalkMatrixElementsError(msg)
+
+    return features.fl_crosstalk.correct_crosstalk(
+        fl1=fl1,
+        fl2=fl2,
+        fl3=fl3,
+        fl_channel=fl_channel,
+        **ctdict)
+
+
+def compute_ctc1(mm):
+    return compute_ctc(mm, fl_channel=1)
+
+
+def compute_ctc2(mm):
+    return compute_ctc(mm, fl_channel=2)
+
+
+def compute_ctc3(mm):
+    return compute_ctc(mm, fl_channel=3)
+
+
+def get_method(fl_channel):
+    if fl_channel == 1:
+        return compute_ctc1
+    elif fl_channel == 2:
+        return compute_ctc2
+    elif fl_channel == 3:
+        return compute_ctc3
+
+
+def register():
+    opts_all = (["fl1_max",
+                 "fl2_max",
+                 "fl3_max"],
+                ["crosstalk fl21",
+                 "crosstalk fl31",
+                 "crosstalk fl12",
+                 "crosstalk fl32",
+                 "crosstalk fl13",
+                 "crosstalk fl23"])
+
+    opts_12 = (["fl1_max",
+                "fl2_max"],
+               ["crosstalk fl21",
+                "crosstalk fl12"])
+
+    opts_13 = (["fl1_max",
+                "fl3_max"],
+               ["crosstalk fl31",
+                "crosstalk fl13"])
+
+    opts_23 = (["fl2_max",
+                "fl3_max"],
+               ["crosstalk fl32",
+                "crosstalk fl23"])
+
+    for flch in [1, 2, 3]:
+        AncillaryFeature(feature_name="fl{}_max_ctc".format(flch),
+                         method=get_method(flch),
+                         req_features=opts_all[0],
+                         req_config=[["calculation", opts_all[1]]],
+                         priority=1)
+
+    for flch in [1, 2]:
+        AncillaryFeature(feature_name="fl{}_max_ctc".format(flch),
+                         method=get_method(flch),
+                         req_features=opts_12[0],
+                         req_config=[["calculation", opts_12[1]]],
+                         priority=0)
+
+    for flch in [1, 3]:
+        AncillaryFeature(feature_name="fl{}_max_ctc".format(flch),
+                         method=get_method(flch),
+                         req_features=opts_13[0],
+                         req_config=[["calculation", opts_13[1]]],
+                         priority=0)
+
+    for flch in [2, 3]:
+        AncillaryFeature(feature_name="fl{}_max_ctc".format(flch),
+                         method=get_method(flch),
+                         req_features=opts_23[0],
+                         req_config=[["calculation", opts_23[1]]],
+                         priority=0)

dclab/rtdc_dataset/feat_anc_core/af_image_contour.py

@@ -0,0 +1,113 @@
+from ... import features
+from .ancillary_feature import AncillaryFeature
+
+
+def compute_contour(mm):
+    cont = features.contour.get_contour_lazily(mask=mm["mask"])
+    return cont
+
+
+def compute_bright(mm):
+    bavg, bsd = features.bright.get_bright(
+        mask=mm["mask"],
+        image=mm["image"],
+        ret_data="avg,sd",
+        )
+    return {"bright_avg": bavg, "bright_sd": bsd}
+
+
+def compute_bright_bc(mm):
+    bavg, bsd = features.bright_bc.get_bright_bc(
+        mask=mm["mask"],
+        image=mm["image"],
+        image_bg=mm["image_bg"],
+        bg_off=mm["bg_off"] if "bg_off" in mm else None,
+        ret_data="avg,sd",
+        )
+    return {"bright_bc_avg": bavg, "bright_bc_sd": bsd}
+
+
+def compute_bright_perc(mm):
+    p10, p90 = features.bright_perc.get_bright_perc(
+        mask=mm["mask"],
+        image=mm["image"],
+        image_bg=mm["image_bg"],
+        bg_off=mm["bg_off"] if "bg_off" in mm else None,
+        )
+    return {"bright_perc_10": p10, "bright_perc_90": p90}
+
+
+def compute_inert_ratio_cvx(mm):
+    return features.inert_ratio.get_inert_ratio_cvx(cont=mm["contour"])
+
+
+def compute_inert_ratio_prnc(mm):
+    return features.inert_ratio.get_inert_ratio_prnc(cont=mm["contour"])
+
+
+def compute_inert_ratio_raw(mm):
+    return features.inert_ratio.get_inert_ratio_raw(cont=mm["contour"])
+
+
+def compute_tilt(mm):
+    return features.inert_ratio.get_tilt(cont=mm["contour"])
+
+
+def compute_volume(mm):
+    vol = features.volume.get_volume(
+        cont=mm["contour"],
+        pos_x=mm["pos_x"],
+        pos_y=mm["pos_y"],
+        pix=mm.config["imaging"]["pixel size"])
+    return vol
+
+
+def register():
+    AncillaryFeature(feature_name="contour",
+                     method=compute_contour,
+                     req_features=["mask"])
+
+    AncillaryFeature(feature_name="bright_avg",
+                     method=compute_bright,
+                     req_features=["image", "mask"])
+
+    AncillaryFeature(feature_name="bright_sd",
+                     method=compute_bright,
+                     req_features=["image", "mask"])
+
+    AncillaryFeature(feature_name="bright_bc_avg",
+                     method=compute_bright_bc,
+                     req_features=["image", "image_bg", "mask"])
+
+    AncillaryFeature(feature_name="bright_bc_sd",
+                     method=compute_bright_bc,
+                     req_features=["image", "image_bg", "mask"])
+
+    AncillaryFeature(feature_name="bright_perc_10",
+                     method=compute_bright_perc,
+                     req_features=["image", "image_bg", "mask"])
+
+    AncillaryFeature(feature_name="bright_perc_90",
+                     method=compute_bright_perc,
+                     req_features=["image", "image_bg", "mask"])
+
+    AncillaryFeature(feature_name="inert_ratio_cvx",
+                     method=compute_inert_ratio_cvx,
+                     req_features=["contour"])
+
+    AncillaryFeature(feature_name="inert_ratio_prnc",
+                     method=compute_inert_ratio_prnc,
+                     req_features=["contour"])
+
+    AncillaryFeature(feature_name="inert_ratio_raw",
+                     method=compute_inert_ratio_raw,
+                     req_features=["contour"])
+
+    AncillaryFeature(feature_name="tilt",
+                     method=compute_tilt,
+                     req_features=["contour"])
+
+    AncillaryFeature(feature_name="volume",
+                     method=compute_volume,
+                     req_features=["contour", "pos_x", "pos_y"],
+                     req_config=[["imaging", ["pixel size"]]])

dclab/rtdc_dataset/feat_anc_core/af_ml_class.py

@@ -0,0 +1,102 @@
+
+import numpy as np
+from .ancillary_feature import AncillaryFeature
+
+
+def get_ml_score_names(mm):
+    """Return a list of all ml_score_??? features"""
+    feats = []
+    # We cannot loop over mm.features because of infinite recursions
+    for ft in mm._feature_candidates:
+        if ft.startswith("ml_score_") and ft in mm:
+            feats.append(ft)
+    return sorted(feats)
+
+
+def compute_ml_class(mm, sanity_checks=True):
+    """Compute the most-probable class membership for all events
+
+    Parameters
+    ----------
+    mm: .RTDCBase
+        instance with the `ml_score_???` features
+    sanity_checks: bool
+        set to `False` to not perform sanity checks (checks whether
+        the scores are between 0 and 1)
+
+    Returns
+    -------
+    ml_class: 1D ndarray
+        The most-probable class for each event in `mm`. If no class
+        can be attributed to an event (because the scores are all
+        `np.nan` or `0` for that event), the class `-1` is used.
+
+    Notes
+    -----
+    I initially thought about also checking whether each feature
+    sums to one, but discarded the idea. Let's assume that a classifier
+    does an awful classification and classifies all events in
+    the same way. If the dataset is cropped at some point (e.g.
+    debris or other events), then this bad classifier has an
+    increased probability compared to another classifier which is
+    perfect at picking out one population. The ml_score values
+    should be just in the range of [0, 1]. This also simplifies
+    export to hdf5 and the work with hierarchy children.
+    """
+    feats = get_ml_score_names(mm)
+
+    # the score matrix
+    score_matrix = np.zeros((len(mm), len(feats)), dtype=float)
+
+    for ii, ft in enumerate(feats):
+        if sanity_checks:
+            if np.nanmax(mm[ft]) > 1:
+                raise ValueError("Feature '{}' has values > 1!".format(ft))
+            elif np.nanmin(mm[ft]) < 0:
+                raise ValueError("Feature '{}' has values < 0!".format(ft))
+        score_matrix[:, ii] = mm[ft]
+
+    # Now compute the maximum for each event. The initial idea was to just
+    # use `ml_class = np.nanargmax(score_matrix, axis=1)`. However, here we
+    # run into these problems:
+    # 1. This does not handle All-NaN slices, e.g. all features are `np.nan`
+    #    for an event.
+    # 2. This does not properly handle manually-rated, zero-valued features,
+    #    e.g. in a situation where we have two features, one with `np.nan`
+    #    and one with `0`, we cannot assign the event to either of the two
+    #    classes.
+    # 3. There is no "unclassified" class (this also becomes apparent in
+    #    point 2). We will set all events that cannot be attributed to a
+    #    class to `-1` in `ml_class`.
+
+    # Define unusable entries:
+    unusable = np.logical_or(np.isnan(score_matrix), (score_matrix == 0))
+    where_idx_nan = np.sum(~unusable, axis=1) == 0
+    score_matrix[where_idx_nan, :] = -1
+    ml_class = np.nanargmax(score_matrix, axis=1)
+    ml_class[where_idx_nan] = -1
+    return ml_class
+
+
+def has_ml_scores(mm):
+    """Check whether the dataset has ml_scores defined"""
+    # Return the sorted score names plus Ancillary feature hashes.
+    # This will be used to determine the hash of the ml_class feature,
+    # which is important in case the user replaces an ML feature
+    # with a new one.
+    features = get_ml_score_names(mm)
+    idlist = []
+    for feat in features:
+        # We also hash any other AncillaryFeature that might implement
+        # this ML score. But this use case is basically non-existent and
+        # the performance impact is probably negligible.
+        candidates = AncillaryFeature.get_instances(feat)
+        idlist.append((feat, [c.hash(mm) for c in candidates]))
+    return idlist
+
+
+def register():
+    AncillaryFeature(feature_name="ml_class",
+                     method=compute_ml_class,
+                     req_func=has_ml_scores,
+                     )