def __init__(self, neuron_id, network): # drop out
self.neuron_id = neuron_id
self.neuron_number = neuron_id[1]
self.netinput = None
self.output = None
self.output_links = []
self.network = network
self.original_activation_function = LinearIO() # drop out
self.activation_function = self.original_activation_function
self.probability_of_drop_out = 0.0 # drop out
self.drop_out_activation_function = ConstantOutput() # drop out
def __init__(self, neuron_id, n_inputs, network, activation_function,
weight_init_function, learning_rate_function): # drop out
self.neuron_id = neuron_id
self.neuron_number = neuron_id[1]
self.n_inputs = n_inputs
self.network = network
self.links = [
Link(weight_init_function) for _ in range(0, n_inputs + 1)
] # +1 for bias link
self.output_links = []
self.activation_function = activation_function
self.original_activation_function = activation_function # drop out
self.probability_of_drop_out = 0.0 # drop out
self.drop_out_activation_function = ConstantOutput() # drop out
self.weight_init_function = weight_init_function
self.learning_rate_function = learning_rate_function
self.error = None
self.error_at_input = None
self.netinput = None
self.output = None
#TODO implement weight-sharing from __future__ import division import math, random import numpy as np import matplotlib.pyplot as plt from pandas import Series, DataFrame import pandas as pd from jorg.neuralnet_v1 import \ NeuralNet, Instance, NetworkDataCollector, weight_init_function_random from jorg.activation_classes import SigmoidIO, LinearIO, ConstantOutput, GaussGauss, Gauss, STDNonMonotonicIOFunction sigmoid = SigmoidIO() linear = LinearIO() constant = ConstantOutput() nonmon = STDNonMonotonicIOFunction() drop_out_prob = 0.5 #disable_2nd_output_neuron = True n_hidden_neurons = 3 learning_rate = -0.1 rotation = 0.0 n_trials = 10 max_epochs = 3000 error_criterion = 0.01 hidden_neurons_io_function = sigmoid output_neurons_io_function = sigmoid data_collection_interval = 1000 n_inputs = 2 n_outputs = 1
#TODO implement weight-sharing from __future__ import division import math, random import numpy as np import matplotlib.pyplot as plt from pandas import Series, DataFrame import pandas as pd from jorg.neuralnet_v1 import \ NeuralNet, Instance, NetworkDataCollector, weight_init_function_random from jorg.activation_classes import SigmoidIO, LinearIO, ConstantOutput, GaussGauss, Gauss, STDNonMonotonicIOFunction sigmoid = SigmoidIO() linear = LinearIO() constant = ConstantOutput() nonmon = STDNonMonotonicIOFunction() disable_2nd_output_neuron = False n_hidden_neurons = 2 learning_rate = -0.5 rotation = 0.0 n_trials = 3 max_epochs = 10000 error_criterion = 0.00001 hidden_neurons_io_function = sigmoid output_neurons_io_function = sigmoid data_collection_interval = 1000 n_inputs = 2 n_outputs = 2 n_neurons_for_each_layer = [n_inputs, n_hidden_neurons, n_outputs]
Instance([1.0, 0.0], [1.0, 0.0, 0.0]),
Instance([1.0, 1.0], [1.0, 1.0, 1.0])
]
n_inputs = 2
n_outputs = 3
n_neurons_for_each_layer = [n_inputs, 6, n_outputs]
n_hidden_layers = 0
if len(n_neurons_for_each_layer) > 2:
n_hidden_layers = len(n_neurons_for_each_layer) - 2
# specify neuron transforms, weight initialization, and learning rate functions... per layer eg: [ hidden_layer_1, hidden_layer_2, output_layer ]
sigmoid = SigmoidIO()
linear = LinearIO()
constant = ConstantOutput()
neurons_ios = [None] + [sigmoid] * n_hidden_layers + [sigmoid]
weight_init_functions = [
None
] + [weight_init_function_random] * n_hidden_layers + [
weight_init_function_random
]
learning_rate_functions = [
None
] + [learning_rate_function] * n_hidden_layers + [learning_rate_function]
results = []
for seed_value in range(10):
print "seed = ", seed_value,