Skip to content

tensorflow_keras.md

Latest commit

 

History

History
65 lines (47 loc) · 2.26 KB

tensorflow_keras.md

File metadata and controls

65 lines (47 loc) · 2.26 KB
Raw

Notes and useful command in TensorFlow and Keras

Installation

To use Keras, you must first install TensorFlow. If you are using a virtual environment (which you should), you must deactivate and reactivate the virtual environment after installing both Keras and TensorFlow.

TensorFlow

  • If your system currently only have CUDA 8.0 (check with nvcc --version), the only compatible version of TensorFlow is 1.4. Install the GPU enabled version (inside your virtual environment) by running pip install tensorflow-gpu==1.4
  • See this link for tested source configurations: https://www.tensorflow.org/install/install_sources#common_installation_problems
  • Deactivate and reactivate virtual environment

Keras

  • Inside the virtual environment, pip install keras
  • Deactivate and reactivate virtual environment

To test the installation, just import keras. You should see a message stating that it is using the TensorFlow backend.

Using TensorFlow and Keras in a Jupyter Notebook

When using the GPU version of TensorFlow and Keras, you might get the following error when trying to import the packages:

ImportError: libcudart.so.8.0: cannot open shared object file: No such file or directory jupyter notebook

To fix this, add the following at the top of your Jupyter notebook configuration file (named jupyter_notebook_config.py found in $HOME/.jupyter)

import os
c = get_config()
os.environ['LD_LIBRARY_PATH'] = '/usr/local/cuda-8.0/lib64:usr/local/cuda-8.0/lib64/libcudart.so.8.0'
c.Spawner.env.update('LD_LIBRARY_PATH')

Check to see that GPU is available

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

Switch between GPU and CPU for training

Set GPU to 0 below for no GPUs (e.g. use CPU only). Set to 1 and TensorFlow will automatically use the GPU.

config = tf.ConfigProto(
    device_count = {'GPU': 0}
)
sess = tf.Session(config=config)

Alternatively, you can execute fit in a context manager:

With CPU

import tensorflow as tf
with tf.device('/cpu:0'):
    network.fit(train_images, train_labels, epochs=15, batch_size=128)

With GPU

import tensorflow as tf
with tf.device('/gpu:0'):
    network.fit(train_images, train_labels, epochs=15, batch_size=128)