import dask.array as da
from dask_ml.utils import svd_flip

# Run svd on a short-fat array with a single chunk
x = da.random.random(size=(100, 1000)).rechunk((-1, -1))
u, s, v = da.linalg.svd(x)
u, v = svd_flip(u, v)
u.compute()
...
/opt/conda/lib/python3.7/site-packages/dask/core.py in _execute_task(arg, cache, dsk)
    119         # temporaries by their reference count and can execute certain
    120         # operations in-place.
--> 121         return func(*(_execute_task(a, cache) for a in args))
    122     elif not ishashable(arg):
    123         return arg

/opt/conda/lib/python3.7/site-packages/dask_ml/utils.py in _svd_flip_copy(x, y, u_based_decision)
     31         return skm.svd_flip(x, y, u_based_decision=u_based_decision)
     32     except ValueError:
---> 33         return skm.svd_flip(x.copy(), y.copy(), u_based_decision=u_based_decision)
     34 
     35 

/opt/conda/lib/python3.7/site-packages/sklearn/utils/extmath.py in svd_flip(u, v, u_based_decision)
    533         signs = np.sign(u[max_abs_cols, range(u.shape[1])])
    534         u *= signs
--> 535         v *= signs[:, np.newaxis]
    536     else:
    537         # rows of v, columns of u

ValueError: operands could not be broadcast together with shapes (1000,1000) (100,1) (1000,1000)