pytorch index_ add_ Usage introduction

x.index_add_(0, assi, y)

For and on behalf of y Every vector in , Add to x On different vectors of , Add it to x On which vector of assi.

x It's the addend ,y It's an addend , It turns out to be and . The three parameters of the function are ：

The first parameter 0 Represents the dimension of addition ,

The second parameter assi Represents the sum of items ,

The third parameter y Represents the addend .

Here's the picture , With the whole 0 Of x As an example , stay dim=0 On the use of y Add to x On .

The parameters in this function assi The outermost dimension and y The outermost dimension of must be consistent ,x The additive dimension and y The dimensions of the additive dimension must be consistent . such as ,assi yes [ dim1, ],y yes [ dim1, dim2 ],x yes [ XX, dim2].

In the above example ,assi yes [3,],y yes [3,3],x yes [3,3], Meet dimensional requirements .

So the actual meaning of this function is ：

according to assi The specified , take y The sequential items of are added to x The specified index Up , In this case assi yes [1, 1, 2], That is to say y The zeroth vector of , The first vector , The second vector , Add to respectively x Of index by 1,1,2 Go up the vector of .

The results are as follows ：