models.py 文件源码

def apply_multiple_masked(func, data, args=(), kwargs={}):
    # Data is a sequence of arrays (i.e. X, y pairs for training)

    datastack = []
    dims = []
    flat = []
    for d in data:
        if d.ndim == 2:
            datastack.append(d)
            dims.append(d.shape[1])
            flat.append(False)
        elif d.ndim == 1:
            datastack.append(d[:, np.newaxis])
            dims.append(1)
            flat.append(True)
        else:
            raise RuntimeError("data arrays have to be 1 or 2D arrays")

    # Decorate functions to work on stacked data
    dims = np.cumsum(dims[:-1])  # dont split by last dim

    unstack = lambda catdata: [d.flatten() if f else d for d, f
                               in zip(np.hsplit(catdata, dims), flat)]

    unstackfunc = lambda catdata, *nargs, **nkwargs: \
        func(*chain(unstack(catdata), nargs), **nkwargs)

    return apply_masked(unstackfunc, np.ma.hstack(datastack), args, kwargs)


#
# Static module properties
#

# Add all models available to the learning pipeline here!