nzw0301/keras-examples

Keras example codes

Dependencies

I recommend using anaconda3

• python libraries
  • Keras
  • tensorflow or theano
  • pydot_ng
  • matplotllib
  • daft (only tutorial )
  • gensim (only save and load word embeddings)
• software
  • graphviz

Contents

Keras 1

• Image
  • CNN_MNIST
  • variational auto encoder for CIFAR10
  • Gumbel-softmax with variational auto encoder for MNIST
  • residual CIFAR10
• Text
  • encoder-decoder
  • Skip-Gram : not hieraltical softmax and negative sampling
• Tutorial
  • keras_tutorial : for Machine Learning seminar at Tsukuba (note: 7MB notebook and ja only)

Keras 2

• autoencoder
• MLP_MNIST
• CNN_CIFAR10
• CBoW : Normal softmax
• RNN_binary_classification
• NNLM
• SCRNLM : Structurally Constrained Recurrent Nets Language Model
• fastText : Bi-gram features and softmax
• Skip-Gram&NG : Skip-gram with negative sampling
• prodLDA : Product of experts model’s LDA based on VAE
• sklearn_MLP_MNIST: Sklearn API example for parameter search on MNIST classification

keras-examples's Issues

How to get topic distribution per document?

For each document there should be a topic distribution: in the current example I expect a numpy array of shape (8000, 20).

How this can be calculated?

reimplementation CRNN

Alternative structures for character-level RNNs. Piotr Bojanowski, Armand Joulin, Tomas Mikolov in ICLR 2016 workshop.

sample code for finding topic distribution for a test dataset

Hi, can you provide some sample code for applying this topic model to an arbitrary test dataset to get its topic distribution?

Refactor codes for keras 2

plan

• Change backend from theano to TensorFlow
• Rewrite tutorial notebook

reimplementation denoising autoencoders

Data for tweet2vec notebook

Can you share the data (./data/tweets.tsv) used in tweet2vec.ipynb ?

Add example using Stateful RNN

• comparing speed
• Language model

Add SCRN

Index out of range

Getting index out of range in Cell 8, keras-examples/CNN_MNIST.ipynb

The value of pos is getting computed to 51 and it complains that this value must be between 1 and 50

for img_index, filters in enumerate(first_layer, start=1):

    for filter_index, mat in enumerate(filters):

        pos = (filter_index)*10+img_index

        draw_digit(mat, nb_filters, show_size, pos)

plt.show()

Now first layer has dimensions : (26,26,5)
hence img_index goes from 1 to 26.

filters has dimension (26,5)
filter_index goes from 0 to 25.

when filter_index = 5 and img_index = 1, pos = 51 and exception is raised

about the data

can you please tell me where to down load the "yellow.ja.txt" and "yellow.en.txt" ? I could not find them through internet

reimplementation node2vec

http://www.kdd.org/kdd2016/papers/files/Paper_218.pdf

Problem with CBoW

Hello,
I tried your CBoW program and it did not work. I have the latest versions of Anaconda, Keras, and Tensorflow. My machine is a mac.
1/ When opening the Alice text, Python complains that it is not UTF-8. I had to convert this text to be able to read it.
2/ The shape of the labels is not correct. You can correct it with the statement
labels.append([word])
instead of
labels.append(word)

Thank you for putting your code online.
Pierre