# Keras 2.x & Tensorflow 2.x 安裝疑難雜症
###### tags: `ISLab`, `Keras`, `Tensorflow`, `ML`

- 2020/02/11
- Python: 3.7.6
- Tensorflow: 2.1.0
- Keras: 2.3.1
- CUDA: 10.1.2

## Build the environment

1. Install Keras

pip install keras

2. Install Tensorflow

For GPU and CPU

pip install tensorflow

or Preview build (CPU)

pip install tf-nightly

- Testing

import keras
# Keras would call Tensorflow.

### Help

- Tensorflow is installed, but "ImportError: DLL load failed".
- Install [Microsoft Visual C++ Redistributable](https://support.microsoft.com/zh-tw/help/2977003/the-latest-supported-visual-c-downloads)
- AttributeError: module 'tensorflow' has no attribute 'get_default_graph'
- Tensorflow and keras version not match.

> Ref:
> https://www.tensorflow.org/install/pip
> https://github.com/keras-team/keras

## GPU support

- Requirements
- [NVIDIA® GPU drivers](https://www.nvidia.com/drivers) - CUDA 10.1 requires 418.x or higher.
- [CUDA® Toolkit](https://developer.nvidia.com/cuda-toolkit-archive) - TensorFlow supports CUDA 10.1 (TensorFlow >= 2.1.0)
- [cuDNN SDK](https://developer.nvidia.com/cudnn) - (>= 7.6)
- Testing

tf.config.experimental.list_physical_devices('GPU')
print("Num GPUs Available: ", len(
    tf.config.experimental.list_physical_devices('GPU')))

### Help
- If it shows "cudart64_101.dll" or "cudart64_100.dll" not found,
"cudart64_101.dll" is in CUDA 10.1
"cudart64_100.dll" is [here](https://www.joe0.com/wp-content/uploads/2019/10/cudart64_100.dll_.zip)

> Ref:
> https://www.tensorflow.org/install/gpu
> [https://www.joe0.com](https://www.joe0.com/2019/10/19/how-resolve-tensorflow-2-0-error-could-not-load-dynamic-library-cudart64_100-dll-dlerror-cudart64_100-dll-not-found/)

## Keras Example
- [Link](https://github.com/keras-team/keras/tree/master/examples)