I am trying to use boost::multi_array and want to create a view where I am
lowering the number of dimensions, but I wish the dimension to be reduced
to be decided at runtime.
I have a vector of integers and this vector size is the same as the number
of dimensions. If the vector [1] value is -1, then the first dimension
will have all values (i.e. range ()), otherwise it will be a number
indicating what slice to use.
You can see in case 2 what I have tried and maybe what I am trying to do.
Case 3 and 4 I am hard coding the dimensions where I want all values, which
does work, but I'd like to get rid of the hard-codedness and make this more
generic.
Furthermore, I have a one dimensional array of values which I am
"converting" to a boost::multi_array so I can slice into it with the views.
I won't know the dimensions until I read the data. Is there a better more
generic way of doing this? Something like a pointer to a N-dimensional
structure where I can just use the view?
Thanks in advance for any ideas!
Joey
Here is my code:
{
double ret_val;
typedef boost::multi_array_types::index_range range;
const VariableInfo &vinfo = GetVariable (si._sensor);
SubSensorType local_subsensors (si._subsensors);
assert (local_subsensors.size () == vinfo._sizes.size ());
switch (vinfo._sizes.size ()) {
case 0 : ret_val = vals [0];
break;
case 1 : { // one-d array
ret_val = (vals [whichPosition]);
}
break;
// Now we get into slices, find the subsensor element that has -1, and
// create a slice along that dimension
case 2 : { // two-dimensional array
typedef boost::const_multi_array_ref <double, 2> array_type;
array_type::index_gen indices;
array_type multi_array (&(vals [0]), vinfo._sizes);
/*
This doesn't work!
array_type::const_array_view <1>::type OneDView;
if (local_subsensors [0] == -1)
OneDView = multi_array [indices [range ()]
[local_subsensors [1]]];
else
if (local_subsensors [1] == -1)
OneDView = multi_array [indices [local_subsensors [0]
[range ()]]];
ret_val = (OneDView [whichPosition]);
*/
}
break;
case 3 : { // three-dimensional array
typedef boost::const_multi_array_ref <double, 3> array_type;
array_type::index_gen indices;
array_type multi_array (&(vals [0]), vinfo._sizes);
array_type::const_array_view <1>::type OneDView =
multi_array [indices [local_subsensors [0]]
[local_subsensors [1]]
[range ()]];
ret_val = (OneDView [whichPosition]);
}
break;
case 4 : { // four-dimensional array
typedef boost::const_multi_array_ref <double, 4> array_type;
array_type::index_gen indices;
array_type multi_array (&(vals [0]), vinfo._sizes);
array_type::const_array_view <1>::type OneDView =
multi_array [indices [local_subsensors [0]]
[local_subsensors [1]]
[local_subsensors [2]]
[range ()]];
ret_val = (OneDView [whichPosition]);
}
}
break;
default : assert (vinfo._sizes.size () < 4);
return (ret_val);_______________________________________________Boost-users mailing list