public void Case11_Scalar()
{
var sh = new NDCoordinatesIncrementor(new int[0]);
sh.Index.Should().ContainInOrder(0);
sh.Next().Should().BeNull();
}
/// <summary>
/// Compute the truth value of x1 OR x2 element-wise.
/// </summary>
/// <param name="lhs">Input boolean array.</param>
/// <param name="rhs">Input boolean array.</param>
/// <returns>Returns True if the arrays are equal.</returns>
/// <remarks>https://docs.scipy.org/doc/numpy/reference/generated/numpy.logical_or.html</remarks>
public static unsafe NDArray <bool> logical_or(NDArray lhs, NDArray rhs)
{
#if _REGEN1
if (lhs.typecode != rhs.typecode)
{
throw new NotImplementedException("please make sure operands have the same data type");
}
else if (lhs.typecode == NPTypeCode.Boolean)
{
if (lhs.Shape.IsScalar && rhs.Shape.IsScalar)
{
return(NDArray.Scalar(*(bool *)lhs.Address || *(bool *)rhs.Address).MakeGeneric <bool>());
}
var(BroadcastedLeftShape, BroadcastedRightShape) = DefaultEngine.Broadcast(lhs.Shape, rhs.Shape);
var lhs_address = (bool *)lhs.Address;
var rhs_address = (bool *)rhs.Address;
var ret = new NDArray <bool>(new Shape(BroadcastedLeftShape.dimensions), true);
Shape retShape = ret.Shape;
//iterate
var ret_address = (bool *)ret.Address;
var incr = new NDCoordinatesIncrementor(BroadcastedLeftShape.dimensions); //doesn't matter which side it is.
int[] current = incr.Index;
do
{
*(ret_address + retShape.GetOffset(current)) = (*(lhs_address + BroadcastedLeftShape.GetOffset(current))) || *(rhs_address + BroadcastedRightShape.GetOffset(current));
} while (incr.Next() != null);
return(ret);
}
public void Case1()
{
var shape = new Shape(3, 3, 1, 2);
var sh = new NDCoordinatesIncrementor(ref shape); //NDCoordinatesIncrementor
sh.Index.Should().ContainInOrder(0, 0, 0, 0);
sh.Next().Should().ContainInOrder(0, 0, 0, 1);
sh.Next().Should().ContainInOrder(0, 1, 0, 0);
sh.Next().Should().ContainInOrder(0, 1, 0, 1);
sh.Next().Should().ContainInOrder(0, 2, 0, 0);
sh.Next().Should().ContainInOrder(0, 2, 0, 1);
sh.Next().Should().ContainInOrder(1, 0, 0, 0);
sh.Next().Should().ContainInOrder(1, 0, 0, 1);
sh.Next().Should().ContainInOrder(1, 1, 0, 0);
sh.Next().Should().ContainInOrder(1, 1, 0, 1);
sh.Next().Should().ContainInOrder(1, 2, 0, 0);
sh.Next().Should().ContainInOrder(1, 2, 0, 1);
sh.Next().Should().ContainInOrder(2, 0, 0, 0);
sh.Next().Should().ContainInOrder(2, 0, 0, 1);
sh.Next().Should().ContainInOrder(2, 1, 0, 0);
sh.Next().Should().ContainInOrder(2, 1, 0, 1);
sh.Next().Should().ContainInOrder(2, 2, 0, 0);
sh.Next().Should().ContainInOrder(2, 2, 0, 1);
sh.Next().Should().BeNull();
}
public void Case10_Scalar_2()
{
var shape = Shape.Scalar;
var sh = new NDCoordinatesIncrementor(ref shape);
sh.Index.Should().ContainInOrder(0);
sh.Next().Should().BeNull();
}
public void Case6()
{
var shape = new Shape(1);
var sh = new NDCoordinatesIncrementor(ref shape);
sh.Index.Should().ContainInOrder(0);
sh.Next().Should().BeNull();
}
public override NDArray ReduceArgMax(NDArray arr, int? axis_)
{
//in order to iterate an axis:
//consider arange shaped (1,2,3,4) when we want to summarize axis 1 (2nd dimension which its value is 2)
//the size of the array is [1, 2, n, m] all shapes after 2nd multiplied gives size
//the size of what we need to reduce is the size of the shape of the given axis (shape[axis])
var shape = arr.Shape;
if (shape.IsEmpty)
return arr;
if (shape.IsScalar || (shape.size == 1 && shape.NDim == 1))
return NDArray.Scalar(0);
if (axis_ == null)
return NDArray.Scalar(argmax_elementwise(arr));
var axis = axis_.Value;
while (axis < 0)
axis = arr.ndim + axis; //handle negative axis
if (axis >= arr.ndim)
throw new ArgumentOutOfRangeException(nameof(axis));
if (shape[axis] == 1)
{
//if the given div axis is 1 and can be squeezed out.
return np.squeeze_fast(arr, axis);
}
//handle keepdims
Shape axisedShape = Shape.GetAxis(shape, axis);
//prepare ret
var ret = new NDArray(NPTypeCode.Int32, axisedShape, false);
var iterAxis = new NDCoordinatesAxisIncrementor(ref shape, axis);
var iterRet = new NDCoordinatesIncrementor(ref axisedShape);
var iterIndex = iterRet.Index;
var slices = iterAxis.Slices;
#if _REGEN
#region Compute
switch (arr.GetTypeCode)
{
%foreach supported_numericals,supported_numericals_lowercase%
case NPTypeCode.#1:
{
int at;
do
{
var iter = arr[slices].AsIterator<#2>();
/// <summary>
/// Random values in a given shape.
/// Create an array of the given shape and populate it with random samples from a uniform distribution over [0, 1).
/// </summary>
public NDArray rand(Shape shape)
{
NDArray ret = new NDArray(typeof(double), shape, false);
unsafe
{
var addr = (double *)ret.Address;
var incr = new NDCoordinatesIncrementor(ref shape);
do
{
*(addr + shape.GetOffset(incr.Index)) = randomizer.NextDouble();
} while (incr.Next() != null);
}
return(ret);
}
/// <remarks>https://docs.scipy.org/doc/numpy/reference/generated/numpy.matmul.html</remarks>
public override NDArray Matmul(NDArray lhs, NDArray rhs)
{
if (lhs.Shape.IsScalar || rhs.Shape.IsScalar)
{
throw new InvalidOperationException("Matmul can't handle scalar multiplication, use `*` or `np.dot(..)` instead");
}
//If the first argument is 1-D, it is promoted to a matrix by prepending a 1 to its dimensions. After matrix multiplication the prepended 1 is removed.
if (lhs.ndim == 1 && rhs.ndim == 2)
{
throw new NotSupportedException("Input operand 1 has a mismatch in its core dimension 0, with gufunc signature (n?,k),(k,m?)->(n?,m?)");
}
if (rhs.ndim == 1)
{
rhs = np.expand_dims(rhs, 1);
}
if (lhs.ndim == 2 || rhs.ndim == 2)
{
return(MultiplyMatrix(lhs, rhs));
}
NDArray l = lhs;
NDArray r = rhs;
(l, r) = np.broadcast_arrays(l, r);
var retShape = l.Shape.Clean();
Console.WriteLine(retShape);
Debug.Assert(l.shape[0] == r.shape[0]);
var len = l.size;
var ret = new NDArray(np._FindCommonArrayType(l.typecode, r.typecode), retShape);
var iterShape = new Shape(retShape.dimensions.Take(retShape.dimensions.Length - 2).ToArray());
var incr = new NDCoordinatesIncrementor(ref iterShape);
var index = incr.Index;
//TODO! we need to create IEnumeable<int> on NDCoordinatesIncrementor so we can do something like this:
//TODO! Parallel.ForEach(incr, i => MultiplyMatrix(l[index], r[index], ret[index]));
for (int i = 0; i < len; i++, incr.Next())
{
MultiplyMatrix(l[index], r[index], ret[index]);
}
return(ret);
}
public Array ToMuliDimArray <T>() where T : unmanaged
{
var ret = Arrays.Create(typeof(T), shape);
var iter = this.AsIterator <T>();
var hasNext = iter.HasNext;
var next = iter.MoveNext;
var coorditer = new NDCoordinatesIncrementor(shape);
var indices = coorditer.Index;
while (hasNext())
{
ret.SetValue(next(), indices);
if (coorditer.Next() == null)
{
break;
}
}
return(ret);
}
/// <summary>
/// Test whether all array elements evaluate to True.
/// </summary>
/// <param name="nd"></param>
/// <returns></returns>
public override bool All(NDArray nd)
{
if (nd.GetTypeCode != NPTypeCode.Boolean)
{
throw new NotSupportedException("DefaultEngine.All supports only boolean dtype."); //TODO!
}
unsafe
{
var addr = (bool *)nd.Address;
var shape = nd.Shape;
var incr = new NDCoordinatesIncrementor(shape.dimensions);
do
{
if (!(*(addr + shape.GetOffset(incr.Index))))
{
return(false);
}
} while (incr.Next() != null);
return(true);
}
}
protected void autoresetDefault_Single()
{
if (typeof(TOut) == typeof(Single))
{
autoresetDefault_NoCast();
return;
}
var localBlock = Block;
Shape shape = Shape;
var convert = Converts.FindConverter <Single, TOut>();
if (!Shape.IsContiguous || Shape.ModifiedStrides)
{
//Shape is sliced, auto-resetting
switch (Type)
{
case IteratorType.Scalar:
{
var offset = shape.TransformOffset(0);
if (offset != 0)
{
MoveNext = () => convert(*((Single *)localBlock.Address + offset));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
}
else
{
MoveNext = () => convert(*((Single *)localBlock.Address));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
}
Reset = () => { };
HasNext = () => true;
break;
}
case IteratorType.Vector:
{
var size = Shape.size;
MoveNext = () =>
{
var ret = convert(*((Single *)localBlock.Address + shape.GetOffset(index++)));
if (index >= size)
{
index = 0;
}
return(ret);
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => index = 0;
HasNext = () => true;
break;
}
case IteratorType.Matrix:
case IteratorType.Tensor:
{
var iterator = new NDCoordinatesIncrementor(ref shape, incr => incr.Reset());
var index = iterator.Index;
Func <int[], int> getOffset = shape.GetOffset;
MoveNext = () =>
{
var ret = convert(*((Single *)localBlock.Address + getOffset(index)));
iterator.Next();
return(ret);
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => iterator.Reset();
HasNext = () => true;
break;
}
default:
throw new ArgumentOutOfRangeException();
}
}
else
{
//Shape is not sliced, auto-resetting
switch (Type)
{
case IteratorType.Scalar:
MoveNext = () => convert(*(Single *)localBlock.Address);
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => { };
HasNext = () => true;
break;
case IteratorType.Vector:
var size = Shape.size;
MoveNext = () =>
{
var ret = convert(*((Single *)localBlock.Address + index++));
if (index >= size)
{
index = 0;
}
return(ret);
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => index = 0;
HasNext = () => true;
break;
case IteratorType.Matrix:
case IteratorType.Tensor:
var iterator = new NDOffsetIncrementorAutoresetting(Shape); //we do not copy the dimensions because there is not risk for the iterator's shape to change.
MoveNext = () => convert(*((Single *)localBlock.Address + iterator.Next()));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
HasNext = () => true;
break;
default:
throw new ArgumentOutOfRangeException();
}
}
}
protected void setDefaults_UInt32() //UInt32 is the input type
{
if (AutoReset)
{
autoresetDefault_UInt32();
return;
}
if (typeof(TOut) == typeof(UInt32))
{
setDefaults_NoCast();
return;
}
var convert = Converts.FindConverter <UInt32, TOut>();
//non auto-resetting.
var localBlock = Block;
Shape shape = Shape;
if (Shape.IsSliced || Shape.IsBroadcasted)
{
//Shape is sliced, not auto-resetting
switch (Type)
{
case IteratorType.Scalar:
{
var hasNext = new Reference <bool>(true);
var offset = shape.TransformOffset(0);
if (offset != 0)
{
MoveNext = () =>
{
hasNext.Value = false;
return(convert(*((UInt32 *)localBlock.Address + offset)));
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
}
else
{
MoveNext = () =>
{
hasNext.Value = false;
return(convert(*((UInt32 *)localBlock.Address)));
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
}
Reset = () => hasNext.Value = true;
HasNext = () => hasNext.Value;
break;
}
case IteratorType.Vector:
{
MoveNext = () => convert(*((UInt32 *)localBlock.Address + shape.GetOffset(index++)));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => index = 0;
HasNext = () => index < Shape.size;
break;
}
case IteratorType.Matrix:
case IteratorType.Tensor:
{
var hasNext = new Reference <bool>(true);
var iterator = new NDCoordinatesIncrementor(ref shape, _ => hasNext.Value = false);
Func <int[], int> getOffset = shape.GetOffset;
var index = iterator.Index;
MoveNext = () =>
{
var ret = convert(*((UInt32 *)localBlock.Address + getOffset(index)));
iterator.Next();
return(ret);
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () =>
{
iterator.Reset();
hasNext.Value = true;
};
HasNext = () => hasNext.Value;
break;
}
default:
throw new ArgumentOutOfRangeException();
}
}
else
{
//Shape is not sliced, not auto-resetting
switch (Type)
{
case IteratorType.Scalar:
var hasNext = new Reference <bool>(true);
MoveNext = () =>
{
hasNext.Value = false;
return(convert(*((UInt32 *)localBlock.Address)));
};
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => hasNext.Value = true;
HasNext = () => hasNext.Value;
break;
case IteratorType.Vector:
MoveNext = () => convert(*((UInt32 *)localBlock.Address + index++));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => index = 0;
HasNext = () => index < Shape.size;
break;
case IteratorType.Matrix:
case IteratorType.Tensor:
var iterator = new NDOffsetIncrementor(Shape); //we do not copy the dimensions because there is not risk for the iterator's shape to change.
MoveNext = () => convert(*((UInt32 *)localBlock.Address + iterator.Next()));
MoveNextReference = () => throw new NotSupportedException("Unable to return references during iteration when casting is involved.");
Reset = () => iterator.Reset();
HasNext = () => iterator.HasNext;
break;
default:
throw new ArgumentOutOfRangeException();
}
}
}
public override NDArray ReduceAMin(NDArray arr, int? axis_, bool keepdims = false, NPTypeCode? typeCode = null)
{
//in order to iterate an axis:
//consider arange shaped (1,2,3,4) when we want to summarize axis 1 (2nd dimension which its value is 2)
//the size of the array is [1, 2, n, m] all shapes after 2nd multiplied gives size
//the size of what we need to reduce is the size of the shape of the given axis (shape[axis])
var shape = arr.Shape;
if (shape.IsEmpty)
return arr;
if (shape.IsScalar || shape.size == 1 && shape.dimensions.Length == 1)
{
var r = typeCode.HasValue ? Cast(arr, typeCode.Value, true) : arr.Clone();
if (keepdims)
r.Storage.ExpandDimension(0);
else if (!r.Shape.IsScalar && r.Shape.size == 1 && r.ndim == 1)
r.Storage.Reshape(Shape.Scalar);
return r;
}
if (axis_ == null)
{
var r = NDArray.Scalar(amin_elementwise(arr, typeCode));
if (keepdims)
r.Storage.ExpandDimension(0);
else if (!r.Shape.IsScalar && r.Shape.size == 1 && r.ndim == 1)
r.Storage.Reshape(Shape.Scalar);
return r;
}
var axis = axis_.Value;
while (axis < 0)
axis = arr.ndim + axis; //handle negative axis
if (axis >= arr.ndim)
throw new ArgumentOutOfRangeException(nameof(axis));
if (shape[axis] == 1)
{
//if the given div axis is 1 and can be squeezed out.
if (keepdims)
return new NDArray(arr.Storage.Alias());
return np.squeeze_fast(arr, axis);
}
//handle keepdims
Shape axisedShape = Shape.GetAxis(shape, axis);
//prepare ret
var ret = new NDArray(typeCode ?? arr.GetTypeCode, axisedShape, false);
var iterAxis = new NDCoordinatesAxisIncrementor(ref shape, axis);
var iterRet = new NDCoordinatesIncrementor(ref axisedShape);
var iterIndex = iterRet.Index;
var slices = iterAxis.Slices;
#if _REGEN1
#region Compute
switch (arr.GetTypeCode)
{
%foreach supported_numericals,supported_numericals_lowercase%
case NPTypeCode.#1:
{
switch (ret.GetTypeCode)
{
%foreach supported_numericals,supported_numericals_lowercase,supported_numericals_defaultvals%
case NPTypeCode.#101:
{
do
{
var iter = arr[slices].AsIterator<#2>();
public override NDArray ReduceProduct(NDArray arr, int? axis_, bool keepdims = false, NPTypeCode? typeCode = null)
{
//in order to iterate an axis:
//consider arange shaped (1,2,3,4) when we want to summarize axis 1 (2nd dimension which its value is 2)
//the size of the array is [1, 2, n, m] all shapes after 2nd multiplied gives size
//the size of what we need to reduce is the size of the shape of the given axis (shape[axis])
var shape = arr.Shape;
if (shape.IsEmpty || shape.size==0)
return NDArray.Scalar(1, (typeCode ?? arr.typecode));
if (shape.IsScalar || (shape.size == 1 && shape.NDim == 1))
{
var r = NDArray.Scalar(typeCode.HasValue ? Converts.ChangeType(arr.GetAtIndex(0), typeCode.Value) : arr.GetAtIndex(0));
if (keepdims)
r.Storage.ExpandDimension(0);
return r;
}
if (axis_ == null)
{
var r = NDArray.Scalar(product_elementwise(arr, typeCode));
if (keepdims)
r.Storage.ExpandDimension(0);
else if (!r.Shape.IsScalar && r.Shape.size == 1 && r.ndim == 1)
r.Storage.Reshape(Shape.Scalar);
return r;
}
var axis = axis_.Value;
while (axis < 0)
axis = arr.ndim + axis; //handle negative axis
if (axis >= arr.ndim)
throw new ArgumentOutOfRangeException(nameof(axis));
if (shape[axis] == 1)
{
//if the given div axis is 1 and can be squeezed out.
if (keepdims)
return new NDArray(arr.Storage.Alias());
return np.squeeze_fast(arr, axis);
}
//handle keepdims
Shape axisedShape = Shape.GetAxis(shape, axis);
var retType = typeCode ?? (arr.GetTypeCode.GetAccumulatingType());
//prepare ret
var ret = new NDArray(retType, axisedShape, false);
var iterAxis = new NDCoordinatesAxisIncrementor(ref shape, axis);
var iterRet = new NDCoordinatesIncrementor(ref axisedShape);
var iterIndex = iterRet.Index;
var slices = iterAxis.Slices;
//resolve the accumulator type
#if _REGEN1
#region Compute
switch (arr.GetTypeCode)
{
%foreach supported_numericals,supported_numericals_lowercase%
case NPTypeCode.#1:
{
switch (retType)
{
%foreach supported_numericals,supported_numericals_lowercase,supported_numericals_onevales%
case NPTypeCode.#101:
{
do
{
var slice = arr[slices];
var iter = slice.AsIterator<#2>();
var moveNext = iter.MoveNext;
var hasNext = iter.HasNext;
|#102 sum = #103;
while (hasNext())
sum *= (#102) moveNext();
ret.Set#101(Converts.To#101(sum), iterIndex);
} while (iterAxis.Next() != null && iterRet.Next() != null);
break;
}
