Skip to content

Latest commit

 

History

History
30 lines (23 loc) · 1.46 KB

2_Install_Libraries.md

File metadata and controls

30 lines (23 loc) · 1.46 KB
Raw

II. Install libraries for building NN

1. Install Tensorflow + Keras

for building the neural network we need to install: TensorFlow is an open-source symbolic math software library which is used for machine learning applications such as neural networks. Keras is an open-source neural-network library written in Python which is capable of running on top of TensorFlow. pip install tensorflow
pip install keras

2. Install CUDA (optional)

If you want to train your neural network on GPU check first your GPU capability.
Install the NVIDIA® CUDA® Toolkit for running scripts on a GPU-accelerated system and the NVIDIA CUDA® Deep Neural Network library (cuDNN) which is a GPU-accelerated library of primitives for deep neural networks.

Enable GPU support for tensorflow:
pip install tensorflow-gpu

3. Check devices

Using tensorflow check which devices are installed in you system:

XLA_CPU device: CPU
XLA_GPU device: NVIDIA GTX-1060 6GB
XLA stands for accelerated linear algebra. It's Tensorflow's relatively optimizing compiler that can further speed up ML models.

I tested on nvidia geforce gtx 1060 6Gb
Run the code to train CNN with GPU mode: every epoch on CPU takes ~3 minutes, on GPU ~ 15 sec.

from tensorflow.python.client import device_lib
devices = device_lib.list_local_devices()
print(devices)