CODEWITHSUNDEEP
pythonnumpyarray-broadcasting
Chris J Harris
Chris J Harris

Reputation: 1841

A way to map one array onto another in numpy?

I have a 2-d array and a 1-d array, shown below. What I'd like to do is to fill the blank spaces in the 2-d array with the product of the 2-d and 1-d array - probably simplest to demonstrate below:

all_holdings = np.array([[1, 0, 0, 2, 0],
                         [2, 0, 0, 1, 0]]).astype('float64')
sub_holdings = np.array([0.2, 0.3, 0.5])

For which I'd like the desired result to be:

array([[1. , 0.2, 0.3, 2. , 1. ],
       [2. , 0.4, 0.6, 1. , 0.5]])

i.e., (workings shown here):

array([[1., 1*0.2, 1*0.3, 2, 2*0.5],
       [2., 2*0.2, 2*0.3, 1, 1*0.5]])

Is anybody able to think of a relatively fast, preferably vectorized way to do this? I have to run this calculation repeatedly on a number of 2-d arrays, though always with the blank spaces in the same location on the 2-d array.

Thanks in advance (and afterwards)

Upvotes: 0

Views: 998

Answers (1)

hpaulj
hpaulj

Reputation: 231355

In [76]: all_holdings = np.array([[1, 0, 0, 2, 0], 
    ...:                          [2, 0, 0, 1, 0]]).astype('float64') 
    ...: sub_holdings = np.array([0.2, 0.3, 0.5])

With one level of iteration:

In [77]: idx = np.where(all_holdings[0,:]==0)[0]                                
In [78]: idx                                                                    
Out[78]: array([1, 2, 4])
In [79]: res = all_holdings.copy()                                              
In [80]: for i,j in zip(idx, sub_holdings): 
    ...:     res[:,i] = res[:,i-1]*j 
    ...:                                                                        
In [81]: res                                                                    
Out[81]: 
array([[1.  , 0.2 , 0.06, 2.  , 1.  ],
       [2.  , 0.4 , 0.12, 1.  , 0.5 ]])

Oops that res[:,2] column is wrong; I need to use something other than idx-1.

Now I can visualize the action better. For example, all the new values are:

In [82]: res[:,idx]                                                             
Out[82]: 
array([[0.2 , 0.06, 1.  ],
       [0.4 , 0.12, 0.5 ]])

OK, I need a way of properly pairing each of the idx values with the right nonzero column.

In [84]: jdx = np.where(all_holdings[0,:])[0]                                   
In [85]: jdx                                                                    
Out[85]: array([0, 3])

This doesn't cut it.

But lets assume we have a proper jdx.

In [87]: jdx = np.array([0,0,3])                                                
In [88]: res = all_holdings.copy()                                              
In [89]: for i,j,v in zip(idx,jdx, sub_holdings): 
    ...:     res[:,i] = res[:,j]*v 
    ...:                                                                        
In [90]: res                                                                    
Out[90]: 
array([[1. , 0.2, 0.3, 2. , 1. ],
       [2. , 0.4, 0.6, 1. , 0.5]])
In [91]: res[:,idx]                                                             
Out[91]: 
array([[0.2, 0.3, 1. ],
       [0.4, 0.6, 0.5]])

I get the same values without iteration:

In [92]: all_holdings[:,jdx]*sub_holdings                                       
Out[92]: 
array([[0.2, 0.3, 1. ],
       [0.4, 0.6, 0.5]])

In [94]: res[:,idx] = res[:,jdx] *sub_holdings                                  
In [95]: res                                                                    
Out[95]: 
array([[1. , 0.2, 0.3, 2. , 1. ],
       [2. , 0.4, 0.6, 1. , 0.5]])

So the key to find the right jdx array. I'll that up to you!

Upvotes: 1

Related Questions