@leriomaggio	+ValerioMaggio
valeriomaggio	vmaggio_at_fbk_dot_eu

git clone https://github.com/leriomaggio/deep-learning-keras-tensorflow.git

1. Multi-layer Fully Connected Networks (and the backends)
2. Hidden Layers features and Embeddings
3. Convolutional Networks
4. Hyperparameter Tuning
5. Cutsom Layers
6. Deep CNN and Residual Networks
7. Transfer Learning and Fine Tuning
8. Recursive Neural Networks
9. AutoEncoders
10. Multi-Modal Networks

This tutorial requires the following packages:

• Python version 3.5

  • Python 3.4 should be fine as well
  • likely Python 2.7 would be also fine, but who knows? :P

• numpy version 1.10 or later: http://www.numpy.org/

• scipy version 0.16 or later: http://www.scipy.org/

• matplotlib version 1.4 or later: http://matplotlib.org/

• pandas version 0.16 or later: http://pandas.pydata.org

• scikit-learn version 0.15 or later: http://scikit-learn.org

• keras version 2.0 or later: http://keras.io

• tensorflow version 1.0 or later: https://www.tensorflow.org

• ipython/jupyter version 4.0 or later, with notebook support

(Optional but recommended):

• pyyaml
• hdf5 and h5py (required if you use model saving/loading functions in keras)
• NVIDIA cuDNN if you have NVIDIA GPUs on your machines. https://developer.nvidia.com/rdp/cudnn-download

The easiest way to get (most) these is to use an all-in-one installer such as Anaconda from Continuum. These are available for multiple architectures.

I'm currently running this tutorial with Python 3 on Anaconda

!python --version

Python 3.5.2

In this repository, files to re-create virtual env with conda are provided for Linux and OSX systems, namely deep-learning.yml and deep-learning-osx.yml, respectively.

To re-create the virtual environments (on Linux, for example):

conda env create -f deep-learning.yml

For OSX, just change the filename, accordingly.

NOTE: Read this section only if after pip installing theano, it raises error in enabling the GPU support!

NVIDIA Drivers and CuDNN must be installed and configured before hand. Please refer to the official Theano documentation for further details.

Since version 0.9 Theano introduced the libgpuarray in the stable release (it was previously only available in the development version).

The goal of libgpuarray is (from the documentation) make a common GPU ndarray (n dimensions array) that can be reused by all projects that is as future proof as possible, while keeping it easy to use for simple need/quick test.

Here are some useful tips (hopefully) I came up with to properly install and configure theano on (Ubuntu) Linux with GPU support:

1. [If you're using Anaconda] conda install theano pygpu should be just fine!

Sometimes it is suggested to install pygpu using the conda-forge channel:

conda install -c conda-forge pygpu

2. [Works with both Anaconda Python or Official CPython]

• Install libgpuarray from source: Step-by-step install libgpuarray user library

• Then, install pygpu from source: (in the same source folder) python setup.py build && python setup.py install

• pip install theano.

After Theano is installed:

echo "[global]
device = cuda
floatX = float32

[lib]
cnmem = 1.0" > ~/.theanorc

To date tensorflow comes in two different packages, namely tensorflow and tensorflow-gpu, whether you want to install the framework with CPU-only or GPU support, respectively.

For this reason, tensorflow has not been included in the conda envs and has to be installed separately.

pip install tensorflow

pip install tensorflow-gpu

Note: NVIDIA Drivers and CuDNN must be installed and configured before hand. Please refer to the official Tensorflow documentation for further details.

All the code provided+ in this tutorial can run even if tensorflow is not installed, and so using theano as the (default) backend!

This is exactly the power of Keras!

Therefore, installing tensorflow is not stricly required!

+: Apart from the Theano and Tensorflow example that are very specific. Anyway you can skip executing those parts, of course.

By default, Keras is configured with theano as backend.

If you want to use tensorflow instead, these are the simple steps to follow:

1. Create the keras.json (if it does not exist):

touch $HOME/.keras/keras.json

2. Copy the following content into the file:

{
    "epsilon": 1e-07,
    "backend": "tensorflow",
    "floatx": "float32",
    "image_data_format": "channels_last"
}

3. Verify it is properly configured:

!cat ~/.keras/keras.json

{
	"epsilon": 1e-07,
	"backend": "tensorflow",
	"floatx": "float32",
	"image_data_format": "channels_last"
}

import numpy as np
import scipy as sp
import pandas as pd
import matplotlib.pyplot as plt
import sklearn

import keras

Using TensorFlow backend.

import numpy
print('numpy:', numpy.__version__)

import scipy
print('scipy:', scipy.__version__)

import matplotlib
print('matplotlib:', matplotlib.__version__)

import IPython
print('iPython:', IPython.__version__)

import sklearn
print('scikit-learn:', sklearn.__version__)

numpy: 1.11.1
scipy: 0.18.0
matplotlib: 1.5.2
iPython: 5.1.0
scikit-learn: 0.18

import keras
print('keras: ', keras.__version__)

# optional
import theano
print('Theano: ', theano.__version__)

import tensorflow as tf
print('Tensorflow: ', tf.__version__)

keras:  2.0.2
Theano:  0.9.0
Tensorflow:  1.0.1