docs

aarontuor · Dec 17, 2016 · 42fa5cd · 42fa5cd
1 parent 5bbf1d6
commit 42fa5cd
Show file tree

Hide file tree

Showing 85 changed files with 2,005 additions and 2,709 deletions.
diff --git a/.cache/v/cache/lastfailed b/.cache/v/cache/lastfailed
diff --git a/.idea/antk.iml b/.idea/antk.iml
diff --git a/.idea/misc.xml b/.idea/misc.xml
diff --git a/.idea/workspace.xml b/.idea/workspace.xml
diff --git a/.ipynb_checkpoints/Untitled-checkpoint.ipynb b/.ipynb_checkpoints/Untitled-checkpoint.ipynb
diff --git a/.pypirc b/.pypirc
@@ -0,0 +1,3 @@
+[server-login]
+username:tuora
+password:Trex1000
diff --git a/MNIST_data/t10k-images-idx3-ubyte.gz b/MNIST_data/t10k-images-idx3-ubyte.gz
diff --git a/MNIST_data/t10k-labels-idx1-ubyte.gz b/MNIST_data/t10k-labels-idx1-ubyte.gz
diff --git a/MNIST_data/train-images-idx3-ubyte.gz b/MNIST_data/train-images-idx3-ubyte.gz
diff --git a/MNIST_data/train-labels-idx1-ubyte.gz b/MNIST_data/train-labels-idx1-ubyte.gz
diff --git a/__init__.pyc b/__init__.pyc
diff --git a/antk.egg-info/PKG-INFO b/antk.egg-info/PKG-INFO
diff --git a/antk.egg-info/SOURCES.txt b/antk.egg-info/SOURCES.txt
diff --git a/antk.egg-info/dependency_links.txt b/antk.egg-info/dependency_links.txt
diff --git a/antk.egg-info/not-zip-safe b/antk.egg-info/not-zip-safe
diff --git a/antk.egg-info/requires.txt b/antk.egg-info/requires.txt
diff --git a/antk.egg-info/top_level.txt b/antk.egg-info/top_level.txt
diff --git a/antk/.ipynb_checkpoints/pandas_tutorial-checkpoint.ipynb b/antk/.ipynb_checkpoints/pandas_tutorial-checkpoint.ipynb
diff --git a/antk/core/generic_model.py b/antk/core/generic_model.py
@@ -26,7 +26,7 @@ def get_feed_list(batch, placeholderdict, supplement=None, train=1, debug=False)
     :param supplement: A dictionary of numpy input matrices with keys corresponding to placeholders in placeholderdict, where the row size of the matrices do not correspond to the number of datapoints. For use with input data intended for `embedding_lookup`_.
     :param dropouts: Dropout tensors in graph.
     :param dropout_flag: Whether to use Dropout probabilities for feed forward.
-    :return: A feed dictionary with keys of placeholder tensors and values of numpy matrices, paired by key
+    :return: A feed dictionary with keys of placeholder tensors and values of numpy matrices
     """
     ph, dt = [], []
     datadict = batch.features.copy()

diff --git a/antk/core/generic_model.pyc b/antk/core/generic_model.pyc
diff --git a/antk/core/node_ops.py b/antk/core/node_ops.py
@@ -20,6 +20,7 @@ def fan_scale(initrange, activation, tensor_in):
         initrange *= numpy.sqrt(2.0/float(tensor_in.get_shape().as_list()[1]))
     else:
         initrange *= (1.0/numpy.sqrt(float(tensor_in.get_shape().as_list()[1])))
+    return initrange
 
 
 
@@ -167,7 +168,7 @@ def lookup(dataname=None,  data=None,  indices=None, distribution='uniform',
     :return: tf.nn.embedding_lookup(wghts, indices), wghts, indices
     """
 
-    if type(data) is loader.HotIndex:
+    if type(data) is loader.IndexVector:
         if makeplace:
             indices = tf.placeholder(tf.int32, [None], name=dataname)
         wghts = weights(distribution, [data.dim, shape[1]], initrange=initrange, l2=l2, name=name+'_weights')

diff --git a/antk/core/node_ops.pyc b/antk/core/node_ops.pyc
diff --git a/antk/lib/decorate.py b/antk/lib/decorate.py
@@ -4,6 +4,13 @@
 import collections
 import numpy
 
+def required():
+    def wrapper(f):
+        def wrap(self, *args, **kwargs):
+            return f(self, *args, **kwargs)
+        return wrap
+    return wrapper
+
 def flatten(container):
     for i in container:
         if isinstance(i, (list,tuple)):
@@ -56,7 +63,7 @@ def node_repr(tensor_node):
     return new_function
 
 def act(func):
-    func = node_op(func)
+    #func = node_op(func)
     @functools.wraps(func)
     def new_function(*args, **kwargs):
         tensor_out = func(*args, **kwargs)
@@ -98,7 +105,7 @@ def ph_rep(ph):
     return 'Placeholder("%s", shape=%s, dtype=%r)' % (ph.name, ph.get_shape().as_list(), ph.dtype)
 
 def pholder(func):
-    func = node_op(func)
+    #func = node_op(func)
     @functools.wraps(func)
     def new_function(*args, **kwargs):
         tensor_out = func(*args, **kwargs)
@@ -110,7 +117,7 @@ def ph_rep(ph):
     return new_function
 
 def variable(func):
-    func = node_op(func)
+    #func = node_op(func)
     @functools.wraps(func)
     def new_function(*args, **kwargs):
         tensor_out = func(*args, **kwargs)
@@ -142,7 +149,7 @@ def loss_repr(loss):
 
 def neural_net(func):
     defaults = get_default_args(func)
-    func = node_op(func)
+    #func = node_op(func)
     @functools.wraps(func)
     def new_function(*args, **kwargs):
         if 'activation' not in defaults:

diff --git a/antk/lib/decorate.pyc b/antk/lib/decorate.pyc
diff --git a/antk/models/configs/res_concat.config b/antk/models/configs/res_concat.config
@@ -0,0 +1,5 @@
+out linear(1, True)
+-h1 residual_dnn($layers, activation=$activation, skiplayers=2, bn=$bn, keep_prob=$keep_prob)
+--x concat($concat_size)
+---huser lookup(dataname='user', initrange=$initrange, shape=[None, $uembed])
+---hitem lookup(dataname='item', initrange=$initrange, shape=[None, $iembed])
diff --git a/antk/models/configs/res_concat.config.py b/antk/models/configs/res_concat.config.py
@@ -0,0 +1,5 @@
+out linear(1, True)
+-h1 dnn($layers, activation='tanhlecun', bn=True, keep_prob=.95)
+--x concat(24)
+---huser lookup(dataname='user', initrange=$initrange, shape=[None, $embed])
+---hitem lookup(dataname='item', initrange=$initrange, shape=[None, $embed])
diff --git a/antk/models/dnn_concat_model.py b/antk/models/dnn_concat_model.py
@@ -3,7 +3,7 @@
 from antk.core import generic_model
 
 def dnn_concat(data, configfile,
-            layers=[16, 8],
+            layers=[16, 8, 8],
             activation='tanhlecun',
             initrange=1e-3,
             bn=True,

diff --git a/antk/models/dnn_concat_model.pyc b/antk/models/dnn_concat_model.pyc
diff --git a/antk/models/res_concat.py b/antk/models/res_concat.py
@@ -0,0 +1,57 @@
+import tensorflow as tf
+from antk.core import config
+from antk.core import generic_model
+
+def dnn_concat(data, configfile,
+            layers=[16, 8],
+            activation='tanhlecun',
+            initrange=1e-3,
+            bn=True,
+            keep_prob=.95,
+            concat_size=24,
+            uembed=32,
+            iembed=32,
+            learnrate=.00001,
+            verbose=True,
+            epochs=10,
+            maxbadcount=20,
+            mb=2000,
+            eval_rate=500):
+
+    with tf.name_scope('ant_graph'):
+        ant = config.AntGraph(configfile,
+                                data=data.dev.features,
+                                marker='-',
+                                graph_name='dnn_concat',
+                                variable_bindings={'layers': layers,
+                                                   'activation': activation,
+                                                   'initrange': initrange,
+                                                   'bn': bn,
+                                                   'keep_prob': keep_prob,
+                                                   'concat_size': concat_size,
+                                                   'uembed': uembed,
+                                                   'iembed': iembed,
+                                                   })
+
+    y = ant.tensor_out
+    y_ = tf.placeholder("float", [None, None], name='Target')
+    ant.placeholderdict['ratings'] = y_  # put the new placeholder in the graph for training
+    with tf.name_scope('objective'):
+        objective = tf.reduce_sum(tf.square(y_ - y))
+    with tf.name_scope('dev_rmse'):
+        dev_rmse = tf.sqrt(tf.div(tf.reduce_sum(tf.square(y - y_)), data.dev.num_examples))
+    with tf.name_scope('training'):
+        model = generic_model.Model(objective, ant.placeholderdict,
+                                    mb=mb,
+                                    learnrate=learnrate,
+                                    verbose=verbose,
+                                    maxbadcount=maxbadcount,
+                                    epochs=epochs,
+                                    evaluate=dev_rmse,
+                                    predictions=y,
+                                    model_name='res_concat',
+                                    make_histograms=False,
+                                    save=False,
+                                    tensorboard=False)
+        model.train(data.train, dev=data.dev, eval_schedule=eval_rate)
+    return model
diff --git a/docs/_build/doctest/output.txt b/docs/_build/doctest/output.txt
diff --git a/docs/_build/doctrees/antk.antk.core.doctree b/docs/_build/doctrees/antk.antk.core.doctree
diff --git a/docs/_build/doctrees/antk.antk.datascripts.doctree b/docs/_build/doctrees/antk.antk.datascripts.doctree
diff --git a/docs/_build/doctrees/antk.antk.doctree b/docs/_build/doctrees/antk.antk.doctree
diff --git a/docs/_build/doctrees/antk.antk.lib.doctree b/docs/_build/doctrees/antk.antk.lib.doctree
diff --git a/docs/_build/doctrees/antk.antk.models.doctree b/docs/_build/doctrees/antk.antk.models.doctree
diff --git a/docs/_build/doctrees/antk.doctree b/docs/_build/doctrees/antk.doctree
diff --git a/docs/_build/doctrees/api.doctree b/docs/_build/doctrees/api.doctree
diff --git a/docs/_build/doctrees/config.doctree b/docs/_build/doctrees/config.doctree
diff --git a/docs/_build/doctrees/environment.pickle b/docs/_build/doctrees/environment.pickle
diff --git a/docs/_build/doctrees/generic_model.doctree b/docs/_build/doctrees/generic_model.doctree
diff --git a/docs/_build/doctrees/loader.doctree b/docs/_build/doctrees/loader.doctree
diff --git a/docs/_build/doctrees/models.doctree b/docs/_build/doctrees/models.doctree
diff --git a/docs/_build/doctrees/node_ops.doctree b/docs/_build/doctrees/node_ops.doctree
diff --git a/docs/_build/html/_modules/dnn_concat_model.html b/docs/_build/html/_modules/dnn_concat_model.html
@@ -139,7 +139,7 @@ <h1>Source code for dnn_concat_model</h1><div class="highlight"><pre>
 <span class="kn">from</span> <span class="nn">antk.core</span> <span class="kn">import</span> <span class="n">generic_model</span>
 
 <div class="viewcode-block" id="dnn_concat"><a class="viewcode-back" href="../models.html#dnn_concat_model.dnn_concat">[docs]</a><span class="k">def</span> <span class="nf">dnn_concat</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">configfile</span><span class="p">,</span>
-            <span class="n">layers</span><span class="o">=</span><span class="p">[</span><span class="mi">16</span><span class="p">,</span> <span class="mi">8</span><span class="p">],</span>
+            <span class="n">layers</span><span class="o">=</span><span class="p">[</span><span class="mi">16</span><span class="p">,</span> <span class="mi">8</span><span class="p">,</span> <span class="mi">8</span><span class="p">],</span>
             <span class="n">activation</span><span class="o">=</span><span class="s1">&#39;tanhlecun&#39;</span><span class="p">,</span>
             <span class="n">initrange</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span>
             <span class="n">bn</span><span class="o">=</span><span class="bp">True</span><span class="p">,</span>

diff --git a/docs/_build/html/_modules/generic_model.html b/docs/_build/html/_modules/generic_model.html
@@ -162,7 +162,7 @@ <h1>Source code for generic_model</h1><div class="highlight"><pre>
 <span class="sd">    :param supplement: A dictionary of numpy input matrices with keys corresponding to placeholders in placeholderdict, where the row size of the matrices do not correspond to the number of datapoints. For use with input data intended for `embedding_lookup`_.</span>
 <span class="sd">    :param dropouts: Dropout tensors in graph.</span>
 <span class="sd">    :param dropout_flag: Whether to use Dropout probabilities for feed forward.</span>
-<span class="sd">    :return: A feed dictionary with keys of placeholder tensors and values of numpy matrices, paired by key</span>
+<span class="sd">    :return: A feed dictionary with keys of placeholder tensors and values of numpy matrices</span>
 <span class="sd">    &quot;&quot;&quot;</span>
     <span class="n">ph</span><span class="p">,</span> <span class="n">dt</span> <span class="o">=</span> <span class="p">[],</span> <span class="p">[]</span>
     <span class="n">datadict</span> <span class="o">=</span> <span class="n">batch</span><span class="o">.</span><span class="n">features</span><span class="o">.</span><span class="n">copy</span><span class="p">()</span>