def _evaluation_7(feed_dict, name_last, notebook):
notebook.add_cell(
markdown_cell(
'### Evaluation of the Graph\n'
'This contain the code to evaluate num_steps times'
' the graph. After evaluation, the variation of error'
'is shown on a plot.'))
evaluation_cell = code_cell()
evaluation_cell.add_line(
'error = np.zeros(num_steps) # used for plot')
evaluation_cell.add_line('')
evaluation_cell.add_line('with tf.Session() as sess:')
evaluation_cell.add_line(' sess.run(init) # init variables')
evaluation_cell.add_line(' for step in range(num_steps):')
evaluation_cell.add_line(
' v, _ = sess.run([%s, train], feed_dict={%s})' %
(name_last, feed_dict))
evaluation_cell.add_line(' error[step] = v')
evaluation_cell.add_line('')
evaluation_cell.add_line('# show plot')
evaluation_cell.add_line('plt.title("Optimization")')
evaluation_cell.add_line('plt.xlabel("Steps")')
evaluation_cell.add_line('plt.ylabel("Error")')
evaluation_cell.add_line('')
evaluation_cell.add_line('plt.plot(range(num_steps), error)')
evaluation_cell.add_line('plt.show()')
notebook.add_cell(evaluation_cell)
def _optimizator_6(name_last, notebook):
notebook.add_cell(
markdown_cell('### Optimizator: Gradient descent\n'
'The gradient descent optimizator implements '
'the following operation for each $w$ input variable'
'(tf.Variable):'
'$ w\' = w- \\alpha \\nabla \\varepsilon $'))
optimizator_cell = code_cell()
optimizator_cell.add_line(
'optimizer = tf.train.GradientDescentOptimizer(alpha)')
optimizator_cell.add_line('train = optimizer.minimize(%s)' % name_last)
notebook.add_cell(optimizator_cell)
def _evaluation_7_cg(feed_dict, name_last, notebook):
notebook.add_cell(
markdown_cell('### Evaluation of the Graph\n'
'This contain the code to evaluate one time'
' the graph.'))
evaluation_cell = code_cell()
evaluation_cell.add_line('with tf.Session() as sess:')
evaluation_cell.add_line(' sess.run(init) # init variables')
evaluation_cell.add_line(' v = sess.run(%s, feed_dict={%s})' %
(name_last, feed_dict))
evaluation_cell.add_line(' print("Value:\\n", v)')
notebook.add_cell(evaluation_cell)
def _inputs_declaration_4(graph, input_op, notebook, visitor):
notebook.add_cell(
markdown_cell(
'### Input nodes\n'
'In tis cell, the entries nodes to the graph are defined'))
inputs_cell = code_cell()
for i_n in input_op:
inputs_cell.add_line(i_n)
inputs_cell.add_line('')
for weigth_node in graph.weights():
name, op = weigth_node.visit(visitor)
inputs_cell.add_line('%s = %s' % (name, op))
notebook.add_cell(inputs_cell)
def _grpah_declaration_5(graph, notebook, visitor):
notebook.add_cell(
markdown_cell('### Graph declaration\n'
'Then, all nodes in the graph are declarated.'))
notebook.add_cell(
code_cell("def create_mul(v0, v1, name=''):\n"
" try:\n"
" return tf.matmul(v0, v1, name=name)\n"
" except(ValueError, TypeError):\n"
" return tf.multiply(v0, v1, name)\n"))
graph_cell = code_cell()
hidden_node_list = graph.hiddenNodes()
for node in hidden_node_list[:-1]:
name, op = node.visit(visitor)
graph_cell.add_line('%s = %s' % (name, op))
name_last, op_last = hidden_node_list[-1].visit(visitor)
graph_cell.add_line('%s = %s' % (name_last, op_last))
graph_cell.add_line('')
graph_cell.add_line('init = tf.global_variables_initializer() '
'# variables should be inicializate')
notebook.add_cell(graph_cell)
return name_last
def _inputs_values_3(alpha, graph, notebook, visitor, is_neural_net):
notebook.add_cell(
markdown_cell('### Inputs Values\n'
'This cell contain the '
'basic configuration for the graph and the '
'values for the input nodes.'))
inputs_values_cell = code_cell()
inputs_values_cell.add_line('# graph configuration')
inputs_values_cell.add_line('alpha = {}'.format(alpha))
inputs_values_cell.add_line('num_steps = 1000')
inputs_values_cell.add_line(
'\n# input values for placeholders (input nodes)')
input_op = []
feed_dict = ''
for input_node in graph.inputs():
value = input_node.get_node().getValue()
name, op = input_node.visit(visitor)
feed_dict += '%s: %s_value, ' % (name, name)
input_op.append('%s = %s' % (name, op))
inputs_values_cell.add_line('{}_value = {}'.format(
name, _format_input_value(value, is_neural_net)))
feed_dict = feed_dict[:-2]
notebook.add_cell(inputs_values_cell)
return feed_dict, input_op
def _intro_1(notebook):
notebook.add_cell(
markdown_cell('# PYHOTAD \n'
'This document represent the '
'tensorflow code of the graph created in'
'the PyHotAD application.'))