"corrected rando

m state handling. Added a new html theme for documentation. Modified default
columns and rows value in NeuralMap constructor"

docs/source/conf.py

@@ -50,7 +50,15 @@
 # The theme to use for HTML and HTML Help pages. See the documentation for
 # a list of builtin themes.
 #
-html_theme = 'alabaster'
+# html_theme = 'alabaster'
+html_theme = "sphinx_rtd_theme"
+
+html_theme_options = {
+ "navigation_depth": 3,
+ "logo_only": True,
+}
+
+html_logo = "logo.png"
 
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,

neural_map/_neural_map.py

@@ -103,8 +103,8 @@ class NeuralMap:
  def __init__(self,
  variables,
  metric,
- columns=20,
- rows=20,
+ columns=10,
+ rows=10,
  hexagonal=True,
  toroidal=False,
  relative_positions=None,
@@ -147,7 +147,6 @@ def distance(first_array, second_array):
  self.width = self.columns
  self.height = self.rows
  self.seed = seed
- random.seed(self.seed)
 
  if weights is not None:
  _check_inputs.ndarray_and_shape(weights, (self.columns, self.rows, self.variables))
@@ -181,6 +180,7 @@ def distance(first_array, second_array):
  self.current_epoch = None
  self._unified_distance_matrix_cache = None
  self._hdbscan_cache = [(None, None, None)] * self.columns * self.rows
+ random.seed(self.seed)
 
  def pca_weights_init(self, data):
  """
@@ -190,6 +190,7 @@ def pca_weights_init(self, data):
  and use the same normalization for the training data.
 
  """
+ random.seed(self.seed)
  principal_components_length, principal_components = eig(cov(transpose(data)))
  components_order = argsort(-principal_components_length)
 
@@ -204,20 +205,23 @@ def uniform_weights_init(self):
  Randomly initialize the weights with values between 0 and 1.
 
  """
+ random.seed(self.seed)
  self.weights = random.rand(self.columns, self.rows, self.variables)
 
  def standard_weights_init(self):
  """
  Randomly initialize the weights with a random distribution with mean of 0 and std of 1.
 
  """
+ random.seed(self.seed)
  self.weights = random.normal(0., 1., (self.columns, self.rows, self.variables))
 
  def pick_from_data_weights_init(self, data):
  """
  Initialize the weights picking random samples from the input data.
 
  """
+ random.seed(self.seed)
  indices = arange(data.shape[0])
 
  for i in range(self.columns):
@@ -228,6 +232,7 @@ def train(self,
  data,
  eval_data=None,
  n_epochs=100,
+ initial_epoch=0,
  weights_init_method='standard',
  initial_learning_rate=1.0,
  final_learning_rate=0.01,
@@ -254,6 +259,10 @@ def train(self,
  n_epochs: int (optional, default 100)
  Number of epochs to train.
  In each epoch, all observations are processed exactly once, in a random sequence.
+ initial_epoch: int (optional, default 0)
+ Initial training epoch.
+ This might be useful when multiple training sessions are applied
+ over the same instance.
  weights_init_method: string (optional, default 'standard')
  Method to initialize weights. It should be one of the followings:
  * pca_weights_init
@@ -387,6 +396,7 @@ def train(self,
 
  _check_inputs.value_type(n_epochs, int)
  _check_inputs.positive(n_epochs)
+ _check_inputs.value_type(initial_epoch, int)
  _check_inputs.value_type(initial_learning_rate, float)
  _check_inputs.positive(initial_learning_rate)
  _check_inputs.value_type(final_learning_rate, float)
@@ -417,12 +427,6 @@ def train(self,
  epochs_quantization_error = zeros(n_epochs)
  epochs_topographic_error = zeros(n_epochs)
 
- # generate observation indices to iterate over
- indices = arange(len(data))
-
- # set random state
- random.seed(self.seed)
-
  # the next variables before the for loop will be used only in case of toroidal topology
 
  # array with map dimensions, do not confuse with the number of columns and rows
@@ -439,11 +443,14 @@ def train(self,
  correction = zeros(self.rows, dtype='int')
  correction[center[1] % 2:: 2] = (center[1] % 2) * 2 - 1
 
- for epoch in range(n_epochs):
+ for epoch in range(initial_epoch, n_epochs):
 
  # set current epoch
  self.current_epoch = epoch
 
+ # generate observation indices to iterate over
+ indices = arange(len(data))
+
  # shuffles all observations
  random.shuffle(indices)
 

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
  name='neural_map',
- version="0.0.4",
+ version="0.0.5",
  author="Franco José Bobadilla",
  author_email='1709673@ucc.edu.ar',
  description='NeuralMap is a data analysis tool based on Self-Organizing Maps',