Numpy ndarray multiplication switching to matrix multiplication

问题

This is the error I am getting:

File "/data/eduardoj/linear.py", line 305, in _fit_model
        de_dl = (dl_dt + de_dt) * dt_dl
      File "/data/eduardoj/MSc-env/lib/python3.4/site-packages/numpy/matrixlib/defmatrix.py", line 343, in __mul__
        return N.dot(self, asmatrix(other))
    ValueError: shapes (1,53097) and (1,53097) not aligned: 53097 (dim 1) != 1 (dim 0)

And this is the piece of code of numpy where it is crashing:

   340     def __mul__(self, other):                                           
>* 341         if isinstance(other, (N.ndarray, list, tuple)) :                
   342             # This promotes 1-D vectors to row vectors                  
   343             return N.dot(self, asmatrix(other))                         
   344         if isscalar(other) or not hasattr(other, '__rmul__') :          
   345             return N.dot(self, other)                                   
   346         return NotImplemented  

(There is a break point there >* )

In my script I have a loop containing the following line:

de_dl = (dloss_dt + dr_dt) * dt_dl

The expected types and shapes for de_dl, dloss_dt, dr_dt and dt_dl are:

ndarray float32 (1, 53097)

So, I just want to compute the element-wise multiplication. I am using pudb3 for debugging my script. I checked that in the first iteration (i == 0) it works great (initially producing zeros). I notice that for this first iteration the thread did NOT reach the break-point I set. In the next iteration (i==1), I decided to stopped right before to calling the multiplication just to make sure the type and shape of dloss_dt, dr_dt and dt_dl were the still the same. They were.

Though they were the same, it seems that the program went through a different set of steps and some how ended in this N.dot multiplication.

So, I am asking for any clue of what might be keeping me from operating just a simple element-wise multiplication.

回答1:

The snippet of numpy source code you showed in your question corresponds to the __mul__ method of np.matrixlib.defmatrix.matrix.

This method gets called when you right-multiply a matrix object with another matrix or array, i.e. A * B is equivalent to A.__mul__(B) if A is a matrix (it doesn't matter whether B is an ndarray or matrix).

The only way that method could possibly be called is if (dl_dt + de_dt) is a matrix (or some other derived matrix class) rather than an ndarray. Therefore, either dl_dt or de_dt is being cast to a matrix somewhere in your code.

Since __mul__ is not called on the first iteration, this must occur somewhere after line 305 in your code (de_dl = (dl_dt + de_dt) * dt_dl).

来源：https://stackoverflow.com/questions/35117660/numpy-ndarray-multiplication-switching-to-matrix-multiplication