N argmax columns out of max values in M rows?

Question

What is the best/shortest way to get the TWO columns of with max value after selecting the THREE max in every specified row

  [[0.  , 0.  , 0.  ],
   [0.19, 0.  , 0.  ],
   [0.  , 0.29, 0.  ],
   [0.42, 0.  , 0.13],
   [0.12, 0.12, 0.13],
   [0.13, 0.1 , 0.26],
   [0.  , 0.  , 0.  ],
   [0.  , 0.12, 0.  ],
   [0.25, 0.  , 0.48],
   [0.  , 0.  , 0.21]])

so the three max vals at rows 3,4,5 are

    In [132]: np.max(ary[[3,4,5],:],axis=1)                                                                                                                                    
    Out[132]: array([0.42, 0.13, 0.26])

now i have to select the columns of the two max values:

  In [133]: np.argmax(ary[[3,4,5],:],axis=1)                                                                                                                                 
  Out[133]: array([0, 2, 2])

in this case that is element[0]=0 and element[2]=2, ignoring element[1]=2

Is there a quicker way of getting the col-ixs of max-of-max ?

there doesnt seem to be direct argmax-max function you have to always do max+argmax (store intermediary result) and do argmax again

---

is this correct :

 np.argsort(np.max(ary[[3,4,5],:],axis=1))[::-1][:2]