data.py 文件源码

def read(self):
        who = matlab.whosmat(self.filename)
        if not who:
            raise IOError("Couldn't load matlab file " + self.filename)
        else:
            ml = matlab.loadmat(self.filename, chars_as_strings=True)

            ml = {a: b for a, b in ml.items() if isinstance(b, np.ndarray)}

            # X is the biggest numeric array
            numarrays = []
            for name, con in ml.items():
                 if issubclass(con.dtype.type, numbers.Number):
                    numarrays.append((name, reduce(lambda x, y: x*y, con.shape, 1)))
            X = None
            if numarrays:
                nameX = max(numarrays, key=lambda x: x[1])[0]
                X = ml.pop(nameX)

            # find an array with compatible shapes
            attributes = []
            if X is not None:
                nameattributes = None
                for name, con in ml.items():
                    if con.shape in [(X.shape[1],), (1, X.shape[1])]:
                        nameattributes = name
                        break
                attributenames = ml.pop(nameattributes).ravel() if nameattributes else range(X.shape[1])
                attributenames = [str(a).strip() for a in attributenames]  # strip because of numpy char array
                attributes = [ContinuousVariable.make(a) for a in attributenames]

            metas = []
            metaattributes = []

            sizemetas = None
            if X is None:
                counts = defaultdict(list)
                for name, con in ml.items():
                    counts[len(con)].append(name)
                if counts:
                    sizemetas = max(counts.keys(), key=lambda x: len(counts[x]))
            else:
                sizemetas = len(X)
            if sizemetas:
                for name, con in ml.items():
                    if len(con) == sizemetas:
                        metas.append(name)

            metadata = []
            for m in sorted(metas):
                f = ml[m]
                metaattributes.append(StringVariable.make(m))
                f.resize(sizemetas, 1)
                metadata.append(f)

            metadata = np.hstack(tuple(metadata))

            domain = Domain(attributes, metas=metaattributes)
            if X is None:
                X = np.zeros((sizemetas, 0))
            return Orange.data.Table.from_numpy(domain, X, Y=None, metas=metadata)