9. High order operation

torch.where（condition,x,y）

Is selected according to conditions A still B
Three parameters ,（condition,A,B）, Generated C It may come from A（ Direct copy ） or B, Part of it comes from A Sure C【0】= A【0】,C【1】=A【1】, But if it comes from C【0,0,1】=A【0,0,1】,C【0,0,2】=B【0,0,2】 etc. , Not as regular as before , You can write condition In the table . In from condition In the table ,1 The place comes from A.

condition The higher the value inside , Taken from the A The higher the probability , Give a condition greater than
0.5 Taken from the A.
for i
for y
if c[i][j]=A[i][j]
This is from cpu The control of the , Very non parallel . We only generate such a condition matrix , Our assignment statements can be highly parallelized , The best thing is to use GPU.

torch.gather（input,dim,index）

Operation of table lookup
Why do you need , Price reference table A, Index of things you need to buy B, Generated C Need basis B According to A Generate items , according to B Index in 0 stay A The collection index in is 0 Item name of .
gather Three parameters , The first is the original table we want to check , The second is dimension , The third is the index table we need , Then output the contents of an original table sorted by the index table .
out[i][j][k]=input[i][j][index[i][j][k] #if dim=2