Solutions to errors in tensorflow operation

 

1.ImportError: DLL load failed: The specified module was not found

 

resolvent ： hold TensorFlow Uninstalled and reinstalled （ Maybe it's because the version doesn't correspond , stay anaconda Delete TensorFlow library , And then install it again ）

2.keras Report errors ：load_weights() got an unexpected keyword arguement 'skip_mmismatch'

The final will be keras upgrade （ My upgrade to 2.1.5 edition ）, This problem is solved .

summary ：

Running keras and tensorflow The program has encountered many problems caused by this version . Because deep learning is developing rapidly , So many frameworks API The update is faster , in the future debug You must pay attention to checking the version problem .

3.pip Report errors ModuleNotFoundError: No module named 'pip._internal'

# wget https://bootstrap.pypa.io/get-pip.py --no-check-certificate

# sudo python get-pip.py

4. Slow download speed , encounter time read out problem

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple xx

5. Import tensorflow：ImportError: libcublas.so.9.0: cannot open shared object file: No such file or director

The reason for this problem is TensorFlow-gpu Version and cuda Caused by mismatch , It is necessary to adjust the reassembly correspondence

see CUDA edition ：

cat /usr/local/cuda/version.txt

see CUDNN edition ：

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

Corresponding table

6. tensorflow.python.framework.errors_impl.NotFoundError: FindFirstFile failed for

This problem usually occurs because the file path is a relative path , Change the path of the problem to the absolute path （ The full path ）, The following example ：

'./VOC2007/Annotations/'   Change it to    'Z:/pycharm_projects/CenterNet_1/VOC2007/Annotations/'