ZHANG Hua

#527

I change the code as follows: ``` #include #include #include using namespace cl::sycl; int main() { buffer valuesBuf { 1024 }; { // Initialize buffer on the host with 0,...

Yes, it is a warning and the result is correct. But I found max is right, but min is wrong. min always reduced to zero. Change value starts from 100:...

what's error in charRNN.cpp ?

charRNN‘s Load/Save is only used as demo for the example. It cannot fit for other examples, nor python predict API except that the net model and parameter names are same...

transpose issue: it is a bug: https://github.com/dmlc/MXNet.cpp/pull/53 and https://github.com/dmlc/MXNet.cpp/pull/54. but, ... even the two issues were resolved, the code still cannot properly run for latest MXNet version after NNVM refactored....

I wanna know how you find file charRNN.cpp since it had been reverted by @lx75249. I cannot see it when clone the repository to local.

如果是推理那应该是数秒出结果（含一次性初始化时间）。如果是训练时每个epoch都会有打印，按照代码本身适配的数据，在典型的GTX1000系列GPU上几十秒到1-2分钟一个epoch。如果搞不清楚状况可以跑一个最简单的全连接网络（multi_layer_perception.cpp），尝试在CPU上跑，或者自己断点debug下，看看你环境哪里出问题了。

examples/MNIST_CNN.cpp这个例子只提供了训练的例子，只做预测还要加一点代码，完成以下事情：读取图片文件，并获取他的图片RGB数据数组，将其写入到一个Tensor中，然后用/p参数跑MNIST_CNN例子即可。 晚点我会在image_io.cpp中增加一个读取BMP格式图片的方法，把这个MNIST_CNN例子补充完整（大概只要加几行代码）就可以了。

已经加好了，命令行参数参考： .\Release\OpenCLNet.exe MNIST_CNN /p :params_file D:\DataSets\MNIST_CNN.clnetparams :file D:\9.bmp 预测图片需使用28*28大小的24位BMP格式保存。黑底白字，和MNIST训练数据集类似。 我的输出结果： [1,@2019-06-04 01:58:34] GeForce GTX 1050 Ti (kernels build: 280ms) [debugger] interactive thread started on device 1. 8 parameters successfully loaded....