Merge pull request #1005 from mila-udem/release-0.2

Release 0.2

.travis.yml

@@ -25,7 +25,7 @@ before_install:
   - # Setup Python environment with BLAS libraries
   - wget -q http://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh -O miniconda.sh
   - chmod +x miniconda.sh
-  - ./miniconda.sh -b
+  - ./miniconda.sh -b -p $HOME/miniconda
   - export PATH=$HOME/miniconda/bin:$PATH
   - conda update -q --yes conda
   - export FUEL_DATA_PATH=$TRAVIS_BUILD_DIR/data
@@ -54,6 +54,7 @@ script:
         cd $TRAVIS_BUILD_DIR
         git clone https://github.com/mila-udem/blocks-examples.git
         cd blocks-examples
+        git checkout e0d7a0e5b60e802634161a63602673717c3e3c78
         nose2 tests
       fi
 after_script:

CONTRIBUTING.rst

@@ -1,3 +1,26 @@
+Filing an issue
+===============
+If you are having a problem, then *before* filing an issue, please verify
+the following:
+
+* That you are using a **compatible version of Python** -- this means version
+  3.4 or newer for mainline Python. Legacy Python support is limited to 2.7 and
+  will eventually be dropped, and not all features may be available; users are
+  encouraged to move to Python 3.x as soon as possible.
+* That you are using **the latest version of Theano** from the GitHub ``master``
+  branch. Blocks is developed concurrently with Theano's bleeding edge development
+  and many problems with using Blocks can be traced to using the latest stable
+  version of Theano (or an insufficiently recent GitHub checkout). Please see the
+  `Blocks installation instructions`_ for more details.
+* You are using the latest Blocks (and Fuel_) from the GitHub ``master``
+  branch. If you are using ``stable``, then if possible, please check if your
+  problem persists if you switch to using ``master``. It may still be worth
+  filing the issue if your problem is fixed in ``master``, if it is a serious
+  enough problem to warrant backporting a fix to ``stable``.
+* That your issue is about the software itself -- either a bug report, feature
+  request, question on how to accomplish a certain defined operation within
+  Blocks, etc. -- and not a general machine learning or neural networks question.
+
 Making a pull request
 =====================
 
@@ -49,6 +72,8 @@ mailing list and the GitHub issues to make sure the answer isn't out there
 already.
 
 .. _Blocks users mailing list: https://groups.google.com/forum/#!forum/blocks-users
+.. _Blocks installation instructions: https://blocks.readthedocs.org/en/latest/setup.html
+.. _Fuel: http://fuel.readthedocs.org/
 .. _quick reference: https://blocks.readthedocs.org/en/latest/development/pull_request.html
 .. _the documentation: https://blocks.readthedocs.org/en/latest/development/index.html#formatting-guidelines
 .. _coding guidelines: https://blocks.readthedocs.org/en/latest/development/index.html#code-guidelines

blocks/__init__.py

@@ -1,5 +1,3 @@
 """The blocks library for parametrized Theano ops."""
-# Scary warning: Adding code to this file can break namespace packages
-# See https://pythonhosted.org/setuptools/setuptools.html#namespace-packages
-__import__("pkg_resources").declare_namespace(__name__)
-__version__ = '0.1.1'
+import blocks.version
+__version__ = blocks.version.version

blocks/algorithms/__init__.py

@@ -14,7 +14,8 @@
 from blocks.graph import ComputationGraph
 from blocks.roles import add_role, ALGORITHM_HYPERPARAMETER, ALGORITHM_BUFFER
 from blocks.theano_expressions import l2_norm
-from blocks.utils import dict_subset, pack, shared_floatx
+from blocks.utils import (dict_subset, pack, shared_floatx,
+                          shared_floatx_zeros_matching)
 
 logger = logging.getLogger(__name__)
 
@@ -420,7 +421,7 @@ def __init__(self, momentum=0.):
         add_role(self.momentum, ALGORITHM_HYPERPARAMETER)
 
     def compute_step(self, parameter, previous_step):
-        velocity = shared_floatx(parameter.get_value() * 0., "velocity")
+        velocity = shared_floatx_zeros_matching(parameter, "velocity")
         add_role(velocity, ALGORITHM_BUFFER)
         step = self.momentum * velocity + previous_step
         updates = [(velocity, step)]
@@ -487,11 +488,11 @@ def __init__(self, decay_rate=0.95, epsilon=1e-6):
         add_role(self.epsilon, ALGORITHM_HYPERPARAMETER)
 
     def compute_step(self, parameter, previous_step):
-        mean_square_step_tm1 = shared_floatx(parameter.get_value() * 0.,
-                                             "mean_square_step_tm1")
+        mean_square_step_tm1 = shared_floatx_zeros_matching(
+            parameter, "mean_square_step_tm1")
         add_role(mean_square_step_tm1, ALGORITHM_BUFFER)
-        mean_square_delta_x_tm1 = shared_floatx(parameter.get_value() * 0.,
-                                                "mean_square_delta_x_tm1")
+        mean_square_delta_x_tm1 = shared_floatx_zeros_matching(
+            parameter, "mean_square_delta_x_tm1")
         add_role(mean_square_delta_x_tm1, ALGORITHM_BUFFER)
 
         mean_square_step_t = (
@@ -550,8 +551,8 @@ def __init__(self, decay_rate=0.9, max_scaling=1e5):
         self.epsilon = 1. / max_scaling
 
     def compute_step(self, parameter, previous_step):
-        mean_square_step_tm1 = shared_floatx(parameter.get_value() * 0.,
-                                             "mean_square_step_tm1")
+        mean_square_step_tm1 = shared_floatx_zeros_matching(
+            parameter, "mean_square_step_tm1")
         add_role(mean_square_step_tm1, ALGORITHM_BUFFER)
         mean_square_step_t = (
             self.decay_rate * mean_square_step_tm1 +
@@ -742,15 +743,16 @@ class AdaGrad(StepRule):
 
     """
     def __init__(self, learning_rate=0.002, epsilon=1e-6):
-        self.learning_rate = learning_rate
-        self.epsilon = epsilon
+        self.learning_rate = shared_floatx(learning_rate, "learning_rate")
+        self.epsilon = shared_floatx(epsilon, "epsilon")
+        add_role(self.learning_rate, ALGORITHM_HYPERPARAMETER)
+        add_role(self.epsilon, ALGORITHM_HYPERPARAMETER)
 
     def compute_step(self, parameter, previous_step):
         name = 'adagrad_sqs'
         if parameter.name:
             name += '_' + parameter.name
-        ssq = shared_floatx(parameter.get_value() * 0.,
-                            name=name)
+        ssq = shared_floatx_zeros_matching(parameter, name=name)
         add_role(ssq, ALGORITHM_BUFFER)
 
         ssq_t = (tensor.sqr(previous_step) + ssq)
@@ -789,16 +791,19 @@ class Adam(StepRule):
     def __init__(self, learning_rate=0.002,
                  beta1=0.1, beta2=0.001, epsilon=1e-8,
                  decay_factor=(1 - 1e-8)):
-        self.learning_rate = learning_rate
-        self.beta1 = beta1
-        self.beta2 = beta2
-        self.epsilon = epsilon
-        self.decay_factor = decay_factor
+        self.learning_rate = shared_floatx(learning_rate, "learning_rate")
+        self.beta1 = shared_floatx(beta1, "beta1")
+        self.beta2 = shared_floatx(beta2, "beta2")
+        self.epsilon = shared_floatx(epsilon, "epsilon")
+        self.decay_factor = shared_floatx(decay_factor, "decay_factor")
+        for param in [self.learning_rate, self.beta1, self.beta2, self.epsilon,
+                      self.decay_factor]:
+            add_role(param, ALGORITHM_HYPERPARAMETER)
 
     def compute_step(self, parameter, previous_step):
-        mean = shared_floatx(parameter.get_value() * 0., 'mean')
+        mean = shared_floatx_zeros_matching(parameter, 'mean')
         add_role(mean, ALGORITHM_BUFFER)
-        variance = shared_floatx(parameter.get_value() * 0., 'variance')
+        variance = shared_floatx_zeros_matching(parameter, 'variance')
         add_role(variance, ALGORITHM_BUFFER)
         time = shared_floatx(0., 'time')
         add_role(time, ALGORITHM_BUFFER)