Get rid of network module.

neupy/algorithms/__init__.py

@@ -1,3 +1,4 @@
+# Algorithms
 from .gd.base import *
 from .gd.lev_marq import *
 from .gd.quasi_newton import *
@@ -46,3 +47,6 @@
 from .linear.perceptron import *
 
 from .rbm import *
+
+# Other
+from .utils import StopTrainingException

neupy/algorithms/associative/base.py

@@ -3,8 +3,7 @@
 from neupy import init
 from neupy.utils import format_data
 from neupy.core.properties import IntProperty, ParameterProperty, ArrayProperty
-from neupy.network import BaseNetwork
-from neupy.network.utils import step_function
+from neupy.algorithms.base import BaseNetwork
 
 
 __all__ = ('BaseStepAssociative',)
@@ -164,7 +163,7 @@ def init_layers(self):
     def predict(self, input_data):
         input_data = format_data(input_data, is_feature1d=False)
         raw_output = input_data.dot(self.weight) + self.bias
-        return step_function(raw_output)
+        return np.where(raw_output > 0, 1, 0)
 
     def train(self, input_train, *args, **kwargs):
         input_train = format_data(input_train, is_feature1d=False)

neupy/algorithms/associative/oja.py

@@ -2,7 +2,7 @@
 
 from neupy.utils import format_data, NotTrainedException
 from neupy.core.properties import IntProperty, ParameterProperty
-from neupy.network.base import BaseNetwork
+from neupy.algorithms.base import BaseNetwork
 from neupy import init
 
 

neupy/network/base.py → neupy/algorithms/base.py

@@ -13,7 +13,7 @@
                                    Property)
 from .summary_info import SummaryTable, InlineSummary
 from .utils import (iter_until_converge, shuffle, normalize_error,
-                    StopNetworkTraining)
+                    StopTrainingException)
 
 
 __all__ = ('BaseNetwork',)
@@ -435,7 +435,7 @@ def train(self, input_train, target_train=None, input_test=None,
 
                     is_first_iteration = False
 
-                except StopNetworkTraining as err:
+                except StopTrainingException as err:
                     # TODO: This notification breaks table view in terminal.
                     # I need to show it in a different way.
                     logs.message("TRAIN", "Epoch #{} stopped. {}"

neupy/algorithms/competitive/art.py

@@ -5,7 +5,7 @@
 from neupy.utils import format_data
 from neupy.core.properties import (ProperFractionProperty,
                                    IntProperty)
-from neupy.network.base import BaseNetwork
+from neupy.algorithms.base import BaseNetwork
 
 
 __all__ = ('ART1',)

neupy/network/constructor.py → neupy/algorithms/constructor.py

@@ -15,9 +15,9 @@
                                       is_sequential)
 from neupy.helpers import table
 from neupy.core.properties import ChoiceProperty
-from neupy.network import errors
-from .learning import SupervisedLearningMixin
-from .base import BaseNetwork
+from neupy.algorithms.base import BaseNetwork
+from neupy.algorithms.learning import SupervisedLearningMixin
+from .gd import errors
 
 
 __all__ = ('ConstructableNetwork',)

neupy/algorithms/gd/base.py

@@ -10,9 +10,9 @@
 
 from neupy.core.config import Configurable
 from neupy.core.properties import Property, BoundedProperty
-from neupy.network import ConstructableNetwork
 from neupy.utils import as_tuple
-from . import addon_types
+from neupy.algorithms.constructor import ConstructableNetwork
+from neupy.algorithms.gd import addon_types
 
 
 __all__ = ('GradientDescent', 'MinibatchGradientDescent')

neupy/algorithms/gd/lev_marq.py

@@ -5,10 +5,9 @@
 import numpy as np
 
 from neupy.utils import asfloat
-from neupy.network import errors
 from neupy.core.properties import BoundedProperty, ChoiceProperty
 from neupy.algorithms import GradientDescent
-from neupy.algorithms.gd import NoStepSelection
+from neupy.algorithms.gd import NoStepSelection, errors
 from neupy.algorithms.utils import (parameters2vector, iter_parameter_values,
                                     setup_parameter_updates)
 

neupy/algorithms/linear/base.py

@@ -1,7 +1,7 @@
 from neupy.utils import is_list_of_integers
-from neupy.layers.connections import NetworkConnectionError, LayerConnection
-from neupy.network import ConstructableNetwork
 from neupy.layers import Step, Input
+from neupy.layers.connections import NetworkConnectionError, LayerConnection
+from neupy.algorithms.constructor import ConstructableNetwork
 
 
 __all__ = ('BaseLinearNetwork',)

neupy/algorithms/memory/bam.py

@@ -3,8 +3,7 @@
 import numpy as np
 
 from neupy.utils import format_data, NotTrainedException
-from neupy.network.utils import step_function
-from .utils import bin2sign, hopfield_energy
+from .utils import bin2sign, hopfield_energy, step_function
 from .base import DiscreteMemory
 
 

neupy/algorithms/memory/cmac.py

@@ -2,8 +2,8 @@
 
 from neupy.utils import format_data
 from neupy.core.properties import IntProperty
-from neupy.network.learning import SupervisedLearningMixin
-from neupy.network.base import BaseNetwork
+from neupy.algorithms.learning import SupervisedLearningMixin
+from neupy.algorithms.base import BaseNetwork
 
 
 __all__ = ('CMAC',)

neupy/algorithms/memory/discrete_hopfield_network.py

@@ -3,9 +3,8 @@
 import numpy as np
 
 from neupy.utils import format_data
-from neupy.network.utils import step_function
 from neupy.core.properties import Property
-from .utils import bin2sign, hopfield_energy
+from .utils import bin2sign, hopfield_energy, step_function
 from .base import DiscreteMemory
 
 

neupy/algorithms/memory/utils.py

@@ -2,7 +2,7 @@
 from numpy.core.umath_tests import inner1d
 
 
-__all__ = ('bin2sign', 'hopfield_energy')
+__all__ = ('bin2sign', 'hopfield_energy', 'step_function')
 
 
 def bin2sign(matrix):
@@ -43,3 +43,18 @@ def hopfield_energy(weight, input_data, output_data):
         Hopfield energy for specific data and weights.
     """
     return -0.5 * inner1d(input_data.dot(weight), output_data)
+
+
+def step_function(input_value):
+    """
+    Step function.
+
+    Parameters
+    ----------
+    input_value : array-like
+
+    Returns
+    -------
+    array-like
+    """
+    return np.where(input_value > 0, 1, 0)

neupy/algorithms/rbfn/grnn.py

@@ -2,8 +2,8 @@
 
 from neupy.utils import format_data, NotTrainedException
 from neupy.core.properties import BoundedProperty
-from neupy.network.base import BaseNetwork
-from neupy.network.learning import LazyLearningMixin
+from neupy.algorithms.base import BaseNetwork
+from neupy.algorithms.learning import LazyLearningMixin
 from .utils import pdf_between_data
 
 

neupy/algorithms/rbfn/pnn.py

@@ -2,8 +2,8 @@
 
 from neupy.utils import format_data, NotTrainedException
 from neupy.core.properties import BoundedProperty
-from neupy.network.base import BaseNetwork
-from neupy.network.learning import LazyLearningMixin
+from neupy.algorithms.base import BaseNetwork
+from neupy.algorithms.learning import LazyLearningMixin
 from neupy.algorithms.gd.base import MinibatchTrainingMixin
 
 from .utils import pdf_between_data

neupy/algorithms/rbfn/rbf_kmeans.py

@@ -7,7 +7,7 @@
 from neupy.utils import format_data
 from neupy.core.properties import IntProperty
 from neupy.algorithms.gd import NoStepSelection
-from neupy.network.base import BaseNetwork
+from neupy.algorithms.base import BaseNetwork
 
 
 __all__ = ('RBFKMeans',)

neupy/algorithms/rbm.py

@@ -4,8 +4,8 @@
 import numpy as np
 
 from neupy.core.properties import IntProperty, ParameterProperty
-from neupy.network.constructor import BaseAlgorithm
-from neupy.network.base import BaseNetwork
+from neupy.algorithms.base import BaseNetwork
+from neupy.algorithms.constructor import BaseAlgorithm
 from neupy.algorithms.gd.base import (MinibatchTrainingMixin,
                                       average_batch_errors)
 from neupy.layers.base import create_shared_parameter

neupy/algorithms/utils.py

@@ -1,10 +1,19 @@
 from itertools import chain
 
+import numpy as np
 import theano.tensor as T
 
 
-__all__ = ('parameters2vector', 'iter_parameter_values',
-           'setup_parameter_updates')
+__all__ = ('StopTrainingException', 'shuffle', 'parameters2vector',
+           'iter_parameter_values', 'setup_parameter_updates',
+           'iter_until_converge', 'normalize_error')
+
+
+class StopTrainingException(Exception):
+    """
+    Exception that needs to be triggered in case of
+    early training interruption.
+    """
 
 
 def iter_parameter_values(network):
@@ -78,3 +87,97 @@ def setup_parameter_updates(parameters, parameter_update_vector):
         start_position = end_position
 
     return updates
+
+
+def iter_until_converge(network, epsilon, max_epochs):
+    logs = network.logs
+
+    # Trigger first iteration and store first error term
+    yield network.last_epoch
+
+    previous_error = error_delta = network.errors.last()
+    epoch = network.last_epoch
+
+    while error_delta > epsilon:
+        epoch = epoch + 1
+        network.last_epoch += 1
+
+        yield epoch
+
+        last_error = network.errors.last()
+        error_delta = abs(last_error - previous_error)
+        previous_error = last_error
+
+        if epoch >= max_epochs and error_delta > epsilon:
+            logs.message("TRAIN", "Epoch #{} interrupted. Network didn't "
+                                  "converge after {} iterations"
+                                  "".format(epoch, max_epochs))
+            return
+
+    if np.isnan(error_delta) or np.isinf(error_delta):
+        logs.message("TRAIN", "Epoch #{} interrupted. Network error value is "
+                              "NaN or Inf.".format(epoch))
+    else:
+        logs.message("TRAIN", "Epoch #{} interrupted. Network converged."
+                              "".format(epoch))
+
+
+def shuffle(*arrays):
+    """
+    Make a random shuffle for all arrays.
+
+    Parameters
+    ----------
+    *arrays
+        List of arrays that should be shuffled.
+
+    Returns
+    -------
+    list
+        List of arrays that contain shuffeled input data.
+    """
+    if not arrays:
+        return tuple()
+
+    arrays_without_none = [array for array in arrays if array is not None]
+
+    if not arrays_without_none:
+        return arrays
+
+    first = arrays_without_none[0]
+    n_samples = first.shape[0]
+
+    for array in arrays_without_none:
+        if n_samples != array.shape[0]:
+            raise ValueError("Cannot shuffle matrices. All matrices should "
+                             "have the same number of rows")
+
+    indices = np.arange(n_samples)
+    np.random.shuffle(indices)
+
+    arrays = list(arrays)
+    for i, array in enumerate(arrays):
+        arrays[i] = array[indices] if array is not None else None
+
+    if len(arrays) == 1:
+        return arrays[0]
+
+    return arrays
+
+
+def normalize_error(output):
+    """
+    Normalize error output when result is non-scalar.
+
+    Parameters
+    ----------
+    output : array-like
+        Input can be any numpy array or matrix.
+
+    Returns
+    -------
+    int, float, None
+        Return sum of all absolute values.
+    """
+    if output is not None:
+        return np.sum(output)

neupy/datasets/reber.py

@@ -4,7 +4,7 @@
 
 import numpy as np
 
-from neupy.network.utils import shuffle
+from neupy.algorithms.utils import shuffle
 
 
 __all__ = ('make_reber', 'is_valid_by_reber', 'make_reber_classification')

neupy/estimators.py

@@ -1,6 +1,6 @@
 from functools import wraps
 
-from neupy.network import errors
+from neupy.algorithms.gd import errors
 from neupy.utils import format_data, number_type