Ada Generic Averaging Function - ada

I have a function which averages a certain numeric value from an array of records. This value is either a natural or an enumerated type delta. I have it summing up the values correctly but my question is this: how do I get the length of an array into a generic type, so that it can divide both integers and delta type numbers?

On your array-of-records use the 'Length attribute; this has the advantage of always working even if your bounds are somewhat odd, like -18..3, or an enumeration, like cheeses..fruits.
Something like:
Function Average( Input : In Array_of_Records ) Return float is
-- You say you already have a summation function, so...
Sum : Natural:= Summation( Input );
Begin
Return Sum / Input'Length;
End Average;
You may need to convert the numeric types, by saying Float(Sum) or the like, as Ada does no automatic type "promotions."

This has some flaws in it, but is this closer to what you wanted ?
NWS.
with Ada.Text_Io;
procedure Main is
generic
type Element_T is private;
Zero : Element_T;
One : Element_T;
type Vec_T is array (Integer range <>) of Element_T;
with function "+"(Left, Right : in Element_T) return Element_T is <>;
with function "/"(Left, Right : in Element_T) return Element_T is <>;
package Arrayops is
function Sum (Vec : in Vec_T) return Element_T;
function Count (Vec : in Vec_T) return Element_T;
function Average (Vec : in Vec_T) return Element_T;
end Arrayops;
package body Arrayops is
function Sum (Vec : in Vec_T) return Element_T is
S : Element_T := Zero;
begin
for I in Vec'First .. Vec'Last loop
S := S + Vec(I);
end loop;
return S;
end Sum;
function Count (Vec : in Vec_T) return Element_T is
C : Element_T := Zero;
begin
for I in Vec'First .. Vec'Last loop
C := C + One;
end loop;
return C;
end Count;
function Average (Vec : in Vec_T) return Element_T is
S : constant Element_T := Sum (Vec);
Len : constant Element_T := Count (Vec);
begin
return S / Len;
end Average;
end Arrayops;
type Fl_Arr_T is array (Integer range <>) of Float;
package Fl_Arr is new Arrayops (Element_T => Float,
Zero => 0.0,
One => 1.0,
Vec_T => Fl_Arr_T);
type Int_Arr_T is array (Integer range <>) of Integer;
package Int_Arr is new Arrayops (Element_T => Integer,
Zero => 0,
One => 1,
Vec_T => Int_Arr_T);
My_Ints : constant Int_Arr_T (1 .. 5) := (6,7,5,1,2);
My_Floats : constant Fl_Arr_T (1 .. 7) := (6.1,7.2,5.3,1.4,2.5,8.7,9.7);
Int_Sum : constant Integer := Int_Arr.Sum (My_Ints);
Int_Count : constant Integer := Int_Arr.Count (My_Ints);
Int_Avg : constant Integer := Int_Arr.Average (My_Ints);
Float_Sum : constant Float := Fl_Arr.Sum (My_Floats);
Float_Count : constant Float := Fl_Arr.Count (My_Floats);
Float_Avg : constant Float := Fl_Arr.Average (My_Floats);
begin
Ada.Text_Io.Put_Line ("Integers => Sum: " & Integer'Image (Int_Sum) & ", Count: " & Integer'Image (Int_Count) & ", Avg: " & Integer'Image (Int_Avg));
Ada.Text_Io.Put_Line ("Floats => Sum: " & Float'Image (Float_Sum) & ", Count: " & Float'Image (Float_Count) & ", Avg: " & Float'Image (Float_Avg));
end Main;
Result :
Integers => Sum: 21, Count: 5, Avg: 4
Floats => Sum: 4.09000E+01, Count: 7.00000E+00, Avg: 5.84286E+00

Expanding on Shark8 a bit here...
Ada allows you to declare array types as unconstrained. Something like
type Array_of_Records is array (Natural range <>) of My_Record;
Gives you a type that can be used for arrays of records with starting and ending array indices that could be anywhere in the range of Natural.
One of the nifty things I can do with such a type is use it as a subroutine parameter, like so:
function Sum (Vector : in Array_of_Records) return Natural;
OK, so inside that routine, how do I know where the array bounds are? By using attributes, like so:
for index in Vector'first..Vector'last loop
or
for index in Vector'range loop
Of course for this to work, you must pass in a perfectly-sized array to your Sum routine. Supppose that isn't what you have. Suppose you instead have a huge array (kind of a buffer) and not all of the values are valid? Well, you keep track of what are the valid values, and pass in only those by using a slice.
Rec_Buffer : Array_of_Records (1..10_000);
Last_Valid_Rec : Natural := 0;
....
--// Rec_Buffer gets loaded with 2,128 values or something. We pass it into Sum
--// like so:
Ada.Text_IO ("Sum of vector is " &
natural'image(Sum (Rec_Buffer (1..Last_Valid_Rec));
(warning - uncompiled code)

Related

How to use Assert and loop_invariants

Specification:
package PolyPack with SPARK_Mode is
type Vector is array (Natural range <>) of Integer;
function RuleHorner (X: Integer; A : Vector) return Integer
with
Pre => A'Length > 0 and A'Last < Integer'Last;
end PolyPack ;
I want to write body of PolyPack package with Assert and loop_invariants that the gnatprove program can prove my function RuleHorner correctness.
I write my function Horner but I don;t know how put assertions and loop_invariants in this program to prove its corectness :
with Ada.Integer_Text_IO;
package body PolyPack with SPARK_Mode is
function RuleHorner (X: Integer; A : Vector) return Integer is
Y : Integer := 0;
begin
for I in 0 .. A'Length - 1 loop
Y := (Y*X) + A(A'Last - I);
end loop;
return Y;
end RuleHorner ;
end PolyPack ;
gnatprove :
overflow check might fail (e.g. when X = 2 and Y = -2)
overflow check might fail
overflow check are for line Y := (Y*X) + A(A'Last - I);
Can someone help me how remove overflow check with loop_invariants
The analysis is correct. The element type for type Vector is Integer. When X = 2, Y = -2, and A(A'Last - I) is less than Integer'First + 4 an underflow will occur. How do you think this should be handled in your program? Loop invariants will not work here because you cannot prove that an overflow or underflow cannot occur.
Is there a way you can design your types and/or subtypes used within Vector and for variables X and Y to prevent Y from overflowing or underflowing?
I am also curious why you want to ignore the last value in your Vector. Are you trying to walk through the array in reverse? If so simply use the following for loop syntax:
for I in reverse A'Range loop

Is it necessary to wrap shared array data in a protected type?

I am aware that it is generally bad practice (and the ARM probably says that this is undefined behavior), but I am attempting to write a fast text parser containing many floating point numbers and it would be very expensive to wrap the loaded text into a protected type given that the data is examined character by character and may have up to a million floats or pass a slice on the stack.
Is it possible in Ada (GNAT) to "safely" divide up an unprotected array for consumption with multiple tasks given that the array is never written and only read?
As in:
Text : array (1..1_000_000) of Character := ...
begin
Task_1.Initialize (Start_Index => 1, End_Index => 10_000);
Task_2.Initialize (Start_Index => 10_001, End_Index => 20_000);
...
Yes. That is safe because there is no race condition associated with reading the data and there is no temporally overlapping write operation.
For example, the following code uses such a technique to perform parallel addition on an array of integers.
package Parallel_Addition is
type Data_Array is array(Integer range <>) of Integer;
type Data_Access is access all Data_Array;
function Sum(Item : in not null Data_Access) return Integer;
end Parallel_Addition;
package body Parallel_Addition is
---------
-- Sum --
---------
function Sum (Item : in not null Data_Access) return Integer is
task type Adder is
entry Set (Min : Integer; Max : Integer);
entry Report (Value : out Integer);
end Adder;
task body Adder is
Total : Integer := 0;
First : Integer;
Last : Integer;
begin
accept Set (Min : Integer; Max : Integer) do
First := Min;
Last := Max;
end Set;
for I in First .. Last loop
Total := Total + Item (I);
end loop;
accept Report (Value : out Integer) do
Value := Total;
end Report;
end Adder;
A1 : Adder;
A2 : Adder;
R1 : Integer;
R2 : Integer;
Mid : constant Integer := (Item'Length / 2) + Item'First;
begin
A1.Set (Min => Item'First, Max => Mid);
A2.Set (Min => Mid + 1, Max => Item'Last);
A1.Report (R1);
A2.Report (R2);
return R1 + R2;
end Sum;
end Parallel_Addition;
with Parallel_Addition; use Parallel_Addition;
with Ada.Text_IO; use Ada.Text_IO;
with Ada.Calendar; use Ada.Calendar;
procedure Parallel_Addition_Test is
The_Data : Data_Access := new Data_Array (1 .. Integer'Last / 5);
Start : Time;
Stop : Time;
The_Sum : Integer;
begin
The_Data.all := (others => 1);
Start := Clock;
The_Sum := Sum (The_Data);
Stop := Clock;
Put_Line ("The sum is: " & Integer'Image (The_Sum));
Put_Line
("Addition elapsed time is " &
Duration'Image (Stop - Start) &
" seconds.");
Put_Line
("Time per addition operation is " &
Float'Image(Float(Stop - Start) / Float(The_Data'Length)) &
" seconds.");
end Parallel_Addition_Test;

Why is array'First always 1 in this example?

I have a array of Nodes:
type NodeArray is array (Positive range 1 .. 5) of XNode;
The node has some data and an integer ID but thats not important right now.
The way I undestand it is that array'First (excuse the abuse of notation) always points or references the first item in a range or array, not the integer of the range type.
My question is why I always get 1 instead of the first entry in my array.
If you need to see more code I can provide it, I just thought I'd keep my example simple and short.
Your
type NodeArray is array (Positive range 1 .. 5) of XNode;
defines a constrained array type (ALRM 3.6(5)), whose first index will always be 1.
If you want to use one type to create array objects with different index ranges, you need an unconstrained array type (ALRM 3.6(3)) with bounds like Positive range <> (note 1, once you’ve created such an object, its bounds are fixed; note 2, instead of Positive you can use any scalar appropriate to the problem).
with Ada.Text_IO; use Ada.Text_IO;
procedure Zython is
type Unconstrained_Node_Array is array (Positive range <>) of Float;
subtype Constrained_Node_Array is Unconstrained_Node_Array (1 .. 5);
U : Unconstrained_Node_Array (42 .. 44); -- must include the index range
C : Constrained_Node_Array; -- the index range is 1 .. 5
begin
for J in U'Range loop
U (J) := Float (J) * 2.0;
end loop;
Put_Line ("U'First: "
& Positive'Image (U'First)
& ", U (U'First): "
& Float'Image (U (U'First)));
for J in C'Range loop
C (J) := Float (J) * 2.0;
end loop;
Put_Line ("C'First: "
& Positive'Image (C'First)
& ", C (C'First): "
& Float'Image (C (C'First)));
end Zython;

Different scalar range in different cirumstance

How to represent data with complete scalar range in the first state then represent it as zero to one in the next state while using the same memory space?
Any approaches to the problem is appreciated, the example procedures does not have to be like that if solution requires them to change.
Example
Reading values from a file and then normalize it. Float_Array is for raw value with any range that comes directly from file.
Feature_Array is for normalized values.
type Float_Array is array (Integer range <>) of Float;
type Feature is new Float range 0.0 .. 1.0;
type Feature_Array is array (Integer range <>) of Feature;
The first step is to read floats into an Float_Array and finding max and min value.
procedure Read (Name : String; Result : out Float_Array; Last : out Integer; Min : out Float; Max : out Float) is
use Ada.Text_IO;
use Ada.Float_Text_IO;
F : File_Type;
begin
Open (F, In_File, Name);
for I in Result'Range loop
exit when End_Of_File (F);
Get (F, Result (I));
Min := Float'Min (Min, Result (I));
Max := Float'Max (Max, Result (I));
Last := I;
end loop;
Close (F);
end;
Float_Array is just temporarily being used to read and find min max. The next step is to normalize all values.
function Normalize (Value : Float; Min, Max : Float) return Float is
begin
return (Value - Min) / (Max - Min);
end;
procedure Normalize (Min : Float; Max : Float; Scale : Float; Result : in out Float_Array) is
begin
for E of Result loop
E := Normalize (E, Min, Max) * Scale;
end loop;
end;
After normalization I want the values to be represented as Feature_Array.
Bad solution that does no range check.
There is no range check so it is not a proper solution. Scaling the values from one to three does not yield range check error. So at this point there is no point to have Feature_Array if there is no range check.
Last : Integer;
Data : Float_Array (1 .. 100);
Min : Float := Float'First;
Max : Float := Float'Last;
begin
Read ("frequency.lines_of_float", Data, Last, Min, Max);
Normalize (Min, Max, 1.0, Data);
-- Normalize (Min, Max, 3.0, Data);
declare
The_Features : Feature_Array (Data'Range) with Address => Data'Address;
begin
Put (The_Features);
end;
I have tried attribute 'Valid on the array i.e. The_Features'Valid but it only works on scalar types. And using 'Valid for range check will involve extra code.
I think that I finally understand what is needed here. You want to have variable of normalized type and not of Floats. (in case of floats one would have to constantly do array overlays or have 2 variables pointing to the same address).
Last : Integer;
The_Features : Feature_Array (1 .. 100);
Min : Float := Float'First;
Max : Float := Float'Last;
begin
declare
Data : Float_Array (The_Features'Range) with Address => The_Features'Address;
begin
Read ("frequency.lines_of_float", Data, Last, Min, Max);
Normalize (Min, Max, 1.0, Data);
-- Normalize (Min, Max, 3.0, Data);
end;
Put (The_Features);
This should work but keep in mind that you have to ensure that the result of Normalize is valid.
It seems that manual range checking is the way to go. I can't find a way to use Ada range checking automatically.
To manually check an array of float is within a range
This uses Ada 2012 - conditional expressions.
This is needed sometimes when variables depends on address.
A := (for all E of Item (Item'First .. Last) => E'Valid);
Assert (A, "Elements of Item is not within range.");
Code
with Ada.Text_IO;
with Ada.Float_Text_IO;
procedure Main is
type Float_Array is array (Integer range <>) of Float;
type Feature is new Float range 0.0 .. 1.0;
type Feature_Array is array (Integer range <>) of Feature;
procedure Read (Name : String; Result : out Float_Array; Last : in out Integer; Min : in out Float; Max : in out Float) is
use Ada.Text_IO;
use Ada.Float_Text_IO;
F : File_Type;
begin
Open (F, In_File, Name);
loop
exit when End_Of_File (F);
Last := Last + 1;
Get (F, Result (Last));
Skip_Line (F);
Min := Float'Min (Min, Result (Last));
Max := Float'Max (Max, Result (Last));
exit when Last = Result'Last;
end loop;
Close (F);
end;
function Normalize (Value : Float; Min, Max : Float) return Float is ((Value - Min) / (Max - Min));
procedure Normalize (Min : Float; Max : Float; Scale : Float; Result : in out Float_Array) is
begin
for E of Result loop
E := Normalize (E, Min, Max) * Scale;
end loop;
end;
procedure Put (Item : Feature_Array) is
use Ada.Float_Text_IO;
use Ada.Text_IO;
begin
for E of Item loop
Put (Float (E), 3, 3, 0);
New_Line;
end loop;
end;
procedure Put (Item : Float_Array) is
use Ada.Float_Text_IO;
use Ada.Text_IO;
begin
for E of Item loop
Put (E, 3, 3, 0);
New_Line;
end loop;
end;
procedure Read (Item : out Feature_Array; Last : in out Integer) with
Pre => Feature_Array'Component_Size = Float_Array'Component_Size,
Post => (for all E of Item (Item'First .. Last) => E >= 0.0 and E <= 1.0);
procedure Read (Item : out Feature_Array; Last : in out Integer) is
Data : Float_Array (Item'Range) with Address => Item'Address;
Min : Float := Float'Last;
Max : Float := Float'First;
begin
Read ("f.ssv", Data, Last, Min, Max);
Ada.Text_IO.Put_Line ("Before normalization.");
Put (Data (Data'First .. Last));
Normalize (Min, Max, 1.0, Data (Data'First .. Last));
end;
F : Feature_Array (-5 .. 10);
Last : Integer := F'First - 1;
begin
Read (F, Last);
Ada.Text_IO.Put_Line ("After normalization.");
Put (F (F'First .. Last));
end;
f.ssv
0.1
11.0
-3.0
Result
Before normalization.
0.100
11.000
-3.000
After normalization.
0.221
1.000
0.000

GNATprove: "postcondition might fail" in simple function

I want to write a simple function that finds the biggest number in given Integer array. Here is specification:
package Maximum with SPARK_Mode is
type Vector is array(Integer range <>) of Integer;
function Maximum (A : in Vector) return Integer
with
SPARK_Mode,
Pre => A'Length > 0,
Post =>
(for all i in A'Range => A(i) <= Maximum'Result)
and then
(for some i in A'Range => A(i) = Maximum'Result);
end Maximum;
And here is function's body:
package body Maximum with SPARK_Mode is
function Maximum (A : in Vector) return Integer
is
Max : Integer := A (A'First);
begin
if (A'Length = 1) then
return Max;
end if;
for I in A'First + 1 .. A'Last loop
pragma Loop_Invariant
(for all Index in A'First .. I - 1 => Max >= A(Index));
if A (I) > Max then
Max := A (I);
end if;
end loop;
return Max;
end Maximum;
end Maximum;
And when I try to prove this function with SPARK, it says that postcondition might fail. I'm trying to understand this for like 5 hours now and I have no idea why it says so. It's really annoying, this function MUST work. Do you have any idea why SPARK behaves so strange? What is a data example for this function to not fullfil its postcondition? It always returns a value taken directly from given array and it is always maximal.
Your mistake is to make a loop invariant, which is weaker than the postcondition:
Specification:
package Maximum
with SPARK_Mode
is
type Vector is array (Integer range <>) of Integer;
function Maximum (A : in Vector) return Integer
with
Pre => A'Length > 0,
Post => (for all i in A'Range => A(i) <= Maximum'Result)
and
(for some i in A'Range => A(i) = Maximum'Result);
end Maximum;
Implementation:
package body Maximum with SPARK_Mode is
function Maximum (A : in Vector) return Integer
is
Max : Integer := A (A'First);
begin
if (A'Length = 1) then
return Max;
end if;
for K in A'First + 1 .. A'Last loop
pragma Loop_Invariant
((for all I in A'First .. K - 1 => A (I) <= Max)
and
(for some I in A'First .. K - 1 => A (I) = Max));
if A (K) > Max then
Max := A (K);
end if;
end loop;
return Max;
end Maximum;
end Maximum;
Project file:
project Maximum is
for Main use ("maximum");
end Maximum;

Resources