机器学习深度学习框架使用问题汇总

1.使用keras做mnist分类时，运行时GPU报错

错误信息如下：

2019-07-06 10:26:32.949617: I T:srcgithub	ensorflow	ensorflowcoreplatformcpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2019-07-06 10:26:33.125786: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:1392] Found device 0 with properties: 
name: GeForce GTX 960 major: 5 minor: 2 memoryClockRate(GHz): 1.2785
pciBusID: 0000:02:00.0
totalMemory: 4.00GiB freeMemory: 3.33GiB
2019-07-06 10:26:33.125952: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:1471] Adding visible gpu devices: 0
2019-07-06 10:26:33.395215: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-07-06 10:26:33.395314: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:958]      0 
2019-07-06 10:26:33.395375: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:971] 0:   N 
2019-07-06 10:26:33.395553: I T:srcgithub	ensorflow	ensorflowcorecommon_runtimegpugpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3062 MB memory) -> physical GPU (device: 0, name: GeForce GTX 960, pci bus id: 0000:02:00.0, compute capability: 5.2)
2019-07-06 10:26:33.982153: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.982424: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.983696: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.983828: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.984097: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.985419: E T:srcgithub	ensorflow	ensorflowstream_executorcudacuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED
2019-07-06 10:26:33.985513: W T:srcgithub	ensorflow	ensorflowstream_executorstream.cc:2009] attempting to perform BLAS operation using StreamExecutor without BLAS support
Traceback (most recent call last):
  File "C:UsersAdministratorAppDataRoamingPythonPython36site-packages	ensorflowpythonclientsession.py", line 1322, in _do_call
    return fn(*args)
  File "C:UsersAdministratorAppDataRoamingPythonPython36site-packages	ensorflowpythonclientsession.py", line 1307, in _run_fn
    options, feed_dict, fetch_list, target_list, run_metadata)
  File "C:UsersAdministratorAppDataRoamingPythonPython36site-packages	ensorflowpythonclientsession.py", line 1409, in _call_tf_sessionrun
    run_metadata)
tensorflow.python.framework.errors_impl.InternalError: Blas GEMM launch failed : a.shape=(32, 784), b.shape=(784, 32), m=32, n=32, k=784
     [[Node: dense_1/MatMul = MatMul[T=DT_FLOAT, _class=["loc:@training/RMSprop/gradients/dense_1/MatMul_grad/MatMul_1"], transpose_a=false, transpose_b=false, _device="/job:localhost/replica:0/task:0/device:GPU:0"](_arg_dense_1_input_0_2/_29, dense_1/kernel/read)]]
     [[Node: loss/mul/_61 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_361_loss/mul", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

解决办法：

在前面设置tensorflow对GPU内存的分配比例：

# 解决报错GPU运行报错的问题
# 这里导入tf，用来修改tf后端的配置
import tensorflow as tf
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
# 将显存容量调到只会使用30%
config.gpu_options.per_process_gpu_memory_fraction = 0.3
# 使用设置好的配置
set_session(tf.Session(config=config))

2.Matplotlib和Qt5py的兼容问题

在Pycharm中使用matplotlib的时候，当取消了show plots in tool windows选项时，报错：

pycharm This application failed to start because it could not find or load the Qt 
platform plugin "windows"

解决办法：

在系统变量中添加QT_PLUGIN_PATH

3.训练途中出现NaN数值，比如loss和accuracy等

1.一般出现NaN时，是因为有一些非法计算过程，例如log(0)，所以我们要检查是否在计算过程中存在tf.math.log()等函数

如果有的话，可以使用tf.log(tf.clip_by_value(y,1e-8,1.0))

2.可以尝试调整学习率

4.使用tensorflow训练时出现调用cudnn错误

错误信息：

tensorflow/stream_executor/cuda/cuda_driver.cc:406 failed call to cuInit: CUDA_ERROR_UNKNOWN

解决方法：

　　在Nvida官网查看显卡所需驱动版本：https://www.geforce.cn/drivers

下载并安装更新，问题解决。