Merge pull request #44 from oujago/develop

version 0.4.0

CHANGES.rst

@@ -2,6 +2,22 @@ Changelog
 ---------
 
 
+0.4.0 (2017.-06-18)
+~~~~~~~~~~~~~~~~~~~
+
+Version 0.4.0.
+
+* Embedding backward
+* Momentum
+* NesterovMomentum
+* Adagrad
+* RMSprop
+* Adadelta
+* Adam
+* Adamax
+
+
+
 0.3.0 (2017-06-15)
 ~~~~~~~~~~~~~~~~~~
 

MANIFEST.in

@@ -10,10 +10,10 @@ include requirements-dev.txt
 recursive-include tests *.py
 
 # examples
-recursive-include examples *.py
+recursive-include examples  *.py  *.txt  *.label
 
 # documents
-recursive-include docs *.rst conf.py *.css Makefile
+recursive-include docs  *.rst  *.py  fix_rtd.css  Makefile  make.bat
 
 # applications
 recursive-include application *.py *.md *.html *.jpg

README.rst

@@ -22,21 +22,13 @@
 
 .. https://codeclimate.com/github/oujago/NumpyDL/badges/gpa.svg
    :target: https://codeclimate.com/github/oujago/NumpyDL
-   :alt: Code Climate
 
 .. image:: https://codeclimate.com/github/oujago/NumpyDL/badges/issue_count.svg
    :target: https://codeclimate.com/github/oujago/NumpyDL
-   :alt: Issue Count
 
 .. image:: https://img.shields.io/github/issues/oujago/NumpyDL.svg
    :target: https://github.com/oujago/NumpyDL
 
-.. image:: https://img.shields.io/github/forks/oujago/NumpyDL.svg
-   :target: https://github.com/oujago/NumpyDL
-
-.. image:: https://img.shields.io/github/stars/oujago/NumpyDL.svg
-   :target: https://github.com/oujago/NumpyDL
-
 .. image:: https://zenodo.org/badge/83100910.svg
    :target: https://zenodo.org/badge/latestdoi/83100910
 
@@ -67,6 +59,7 @@ Its main features are:
 5. *API* like ``Keras`` library
 6. *Examples* for several AI tasks
 7. *Application* for a toy chatbot
+8. *Mobile friendly* documents
 
 
 Documentation
@@ -105,19 +98,17 @@ Examples
 ``NumpyDL`` provides several examples of AI tasks:
 
 * sentence classification
-    * LSTM in `examples/lstm_sentence_classification.py`
-    * CNN in `examples/cnn_sentence_classification.py`
+    * LSTM in *examples/lstm_sentence_classification.py*
+    * CNN in *examples/cnn_sentence_classification.py*
 * mnist handwritten recognition
-    * MLP in `examples/mlp-mnist.py`
-    * MLP in `examples/mlp-digits.py`
-    * CNN in `examples/cnn-minist.py`
+    * MLP in *examples/mlp-mnist.py*
+    * MLP in *examples/mlp-digits.py*
+    * CNN in *examples/cnn-minist.py*
 * language modeling
-    * RNN in `examples/rnn-character-lm.py`
-    * RNN in `examples/rnn-character-lm2.py`
-    * LSTM in `examples/lstm-character-lm.py`
-    * LSTM in `examples/lstm-character-lm2.py`
+    * RNN in *examples/rnn-character-lm.py*
+    * LSTM in *examples/lstm-character-lm.py*
 
-One concrete code example in `examples/mlp-digits.py`:
+One concrete code example in *examples/mlp-digits.py*:
 
 .. code-block:: python
 
@@ -152,11 +143,68 @@ Applications
 ``NumpyDL`` provides one toy application:
 
 * Chatbot
-    * seq2seq in `applications/chatbot/model.py`
+    * seq2seq in *applications/chatbot/model.py*
 
 
 And its final result:
 
 .. figure:: applications/chatbot/pics/chatbot.png
     :width: 80%
 
+
+Supports
+========
+
+``NumpyDL`` supports following deep learning techniques:
+
+* Layers
+    * Linear
+    * Dense
+    * Softmax
+    * Dropout
+    * Convolution
+    * Embedding
+    * BatchNormal
+    * MeanPooling
+    * MaxPooling
+    * SimpleRNN
+    * GRU
+    * LSTM
+    * Flatten
+    * DimShuffle
+* Optimizers
+    * SGD
+    * Momentum
+    * NesterovMomentum
+    * Adagrad
+    * RMSprop
+    * Adadelta
+    * Adam
+    * Adamax
+* Objectives
+    * MeanSquaredError
+    * HellingerDistance
+    * BinaryCrossEntropy
+    * SoftmaxCategoricalCrossEntropy
+* Initializations
+    * Zero
+    * One
+    * Uniform
+    * Normal
+    * LecunUniform
+    * GlorotUniform
+    * GlorotNormal
+    * HeNormal
+    * HeUniform
+    * Orthogonal
+* Activations
+    * Sigmoid
+    * Tanh
+    * ReLU
+    * Linear
+    * Softmax
+    * Elliot
+    * SymmetricElliot
+    * SoftPlus
+    * SoftSign
+

applications/chatbot/main.py

@@ -24,7 +24,6 @@ def ask():
         if message == "quit":
             exit()
         else:
-            # print bot_response
             return jsonify({'status': 'OK',
                             'answer': model.utter(message)})
 

examples/lstm-character-lm.py

@@ -7,7 +7,9 @@
 import npdl
 
 
-def main(max_iter, corpus_path=os.path.join(os.path.dirname(__file__), 'data/lm/tiny_shakespeare.txt')):
+def get_data():
+    corpus_path = os.path.join(os.path.dirname(__file__), 'data/lm/tiny_shakespeare.txt')
+
     raw_text = open(corpus_path, 'r').read()
     chars = list(set(raw_text))
     data_size, vocab_size = len(raw_text), len(chars)
@@ -16,7 +18,6 @@ def main(max_iter, corpus_path=os.path.join(os.path.dirname(__file__), 'data/lm/
     index_to_char = {i: ch for i, ch in enumerate(chars)}
 
     time_steps, batch_size = 30, 40
-
     length = batch_size * 20
     text_pointers = np.random.randint(data_size - time_steps - 1, size=length)
     batch_in = np.zeros([length, time_steps, vocab_size])
@@ -26,6 +27,12 @@ def main(max_iter, corpus_path=os.path.join(os.path.dirname(__file__), 'data/lm/
         batch_in[i, range(time_steps), b_[:-1]] = 1
         batch_out[i, b_[-1]] = 1
 
+    return batch_size, vocab_size, time_steps, batch_in, batch_out
+
+
+def main1(max_iter):
+    batch_size, vocab_size, time_steps, batch_in, batch_out = get_data()
+
     print("Building model ...")
     net = npdl.Model()
     net.add(npdl.layers.BatchLSTM(n_out=300, n_in=vocab_size, return_sequence=True,
@@ -39,5 +46,21 @@ def main(max_iter, corpus_path=os.path.join(os.path.dirname(__file__), 'data/lm/
     net.fit(batch_in, batch_out, max_iter=max_iter, batch_size=batch_size)
 
 
+def main2(max_iter):
+    batch_size, vocab_size, time_steps, batch_in, batch_out = get_data()
+
+    print("Building model ...")
+    net = npdl.Model()
+    net.add(npdl.layers.BatchLSTM(n_out=300, n_in=vocab_size, return_sequence=False,
+                                  nb_batch=batch_size, nb_seq=time_steps))
+    # net.add(npdl.layers.MeanPooling(pool_size=(time_steps, 1)))
+    # net.add(npdl.layers.Flatten())
+    net.add(npdl.layers.Softmax(n_out=vocab_size))
+    net.compile(loss=npdl.objectives.SCCE(), optimizer=npdl.optimizers.SGD(lr=0.00001, clip=5))
+
+    print("Train model ...")
+    net.fit(batch_in, batch_out, max_iter=max_iter, batch_size=batch_size)
+
+
 if __name__ == '__main__':
-    main(100)
+    main1(100)

examples/lstm_sentence_classification.py

@@ -62,5 +62,24 @@ def main(max_iter):
     net.fit(xs, ys, batch_size=nb_batch, validation_split=0.1, max_iter=max_iter)
 
 
+def main2(max_iter):
+    nb_batch = 30
+    nb_seq = 20
+
+    xs, ys, x_size, y_size = prepare_data(nb_seq)
+
+    net = npdl.Model()
+    net.add(npdl.layers.Embedding(nb_batch=nb_batch, nb_seq=nb_seq,
+                                  n_out=200, input_size=x_size,
+                                  static=False))
+    net.add(npdl.layers.BatchLSTM(n_out=400, return_sequence=True))
+    net.add(npdl.layers.BatchLSTM(n_out=200, return_sequence=True))
+    net.add(npdl.layers.MeanPooling((nb_seq, 1)))
+    net.add(npdl.layers.Flatten())
+    net.add(npdl.layers.Softmax(n_out=y_size))
+    net.compile(loss='scce', optimizer=npdl.optimizers.RMSprop())
+    net.fit(xs, ys, batch_size=nb_batch, validation_split=0.1, max_iter=max_iter)
+
+
 if __name__ == '__main__':
-    main(100)
+    main2(100)