[fix] add data type check to array setter

[KexinFeng]

Description

Fix #1970. The test code is therein.

details

A poential place to convert the datatype from the input data type to the target array data type is in

djl/api/src/main/java/ai/djl/ndarray/BaseNDManager.java

Line 428 in aeb9135

public static void copyBuffer(Buffer src, ByteBuffer target) {

Here it does the job of:

2. Multiple dataType maps to the same Buffer type, we have to handle them differently

But it is not ideal to conver the Buffer type inside java
https://stackoverflow.com/questions/38745123/bytebufferasshortbuffer-cannot-be-cast-to-java-nio-floatbuffer
Cannot cast 'java.nio.HeapIntBuffer' to 'java.nio.FloatBuffer'

So here we simply throw an exception when datatypes don't match.

[frankfliu]

I don't think we check list this:

1. We should always allow ByteBuffer to set for any DataType
2. Multiple dataType maps to the same Buffer type, we have to handle them differently

[KexinFeng]

I find that in the implementation of set(Buffer data) in different engines, like MXNet, PyTorch, TensorFlow, the buffer is already converted to ByteBuffer before being fed into the engines. Do you mean we should add this to NDArrayAdapter?

[KexinFeng]

By check listing those, is it out of the purpose of automatically converting the input datatype into the target array datatype?

[codecov-commenter]

Codecov Report

Merging #1975 (b3beab7) into master (bb5073f) will decrease coverage by 2.27%.
The diff coverage is 67.74%.

@@             Coverage Diff              @@
##             master    #1975      +/-   ##
============================================
- Coverage     72.08%   69.81%   -2.28%     
- Complexity     5126     5899     +773     
============================================
  Files           473      584     +111     
  Lines         21970    26159    +4189     
  Branches       2351     2824     +473     
============================================
+ Hits          15838    18262    +2424     
- Misses         4925     6521    +1596     
- Partials       1207     1376     +169     

Impacted Files	Coverage Δ
api/src/main/java/ai/djl/modality/cv/Image.java	69.23% <ø> (-4.11%)	⬇️
...rc/main/java/ai/djl/modality/cv/MultiBoxPrior.java	76.00% <ø> (ø)
...rc/main/java/ai/djl/modality/cv/output/Joints.java	71.42% <ø> (ø)
.../main/java/ai/djl/modality/cv/output/Landmark.java	100.00% <ø> (ø)
...main/java/ai/djl/modality/cv/output/Rectangle.java	72.41% <0.00%> (ø)
...i/djl/modality/cv/translator/BigGANTranslator.java	21.42% <0.00%> (-5.24%)	⬇️
...odality/cv/translator/BigGANTranslatorFactory.java	33.33% <0.00%> (+8.33%)	⬆️
.../cv/translator/InstanceSegmentationTranslator.java	0.00% <0.00%> (-86.59%)	⬇️
...nslator/InstanceSegmentationTranslatorFactory.java	7.14% <0.00%> (-11.04%)	⬇️
.../cv/translator/SemanticSegmentationTranslator.java	0.00% <0.00%> (ø)
... and 499 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[KexinFeng]

The following mismatch case will escape the check:
type != dataType && buffer is ByteBuffer, but dataType is not one of the byte types = {DataType.UINT8, DataType.INT8, DataType.BOOLEAN}

Suggested change

	if (!isByteBuffer && type != dataType) {
	if (arrayType != inputType ) {
	DataType[] types = {DataType.UINT8, DataType.INT8, DataType.BOOLEAN};
	if (!isByteBuffer || Arrays.stream(types).noneMatch(x -> x == dataTypeType)) {

[frankfliu]

ByteBuffer should always be allowed. It's the memory representation of all data types. All NDArray has toByteBuffer() method. The JNI only accept ByteBuffer, if we only accept matching Buffer, we have to covert ByteBuffer to Buffer and copy the Buffer to ByteBuffer to pass to JNI. It's not efficient for Hybrid engine. We trying to achieve 0 copy between pytorch and onnx engines. It rely on ByteBuffer.