Class: LSTM_NETWORK

Inherits:
Object
  • Object
show all
Defined in:
lib/NETWORK.rb

Overview

██╗░░░░░░██████╗████████╗███╗░░░███╗ ███╗░░██╗███████╗████████╗░██╗░░░░░░░██╗░█████╗░██████╗░██╗░░██╗██║░░░░░██╔════╝╚══██╔══╝████╗░████║ ████╗░██║██╔════╝╚══██╔══╝░██║░░██╗░░██║██╔══██╗██╔══██╗██║░██╔╝██║░░░░░╚█████╗░░░░██║░░░██╔████╔██║ ██╔██╗██║█████╗░░░░░██║░░░░╚██╗████╗██╔╝██║░░██║██████╔╝█████═╝░██║░░░░░░╚═══██╗░░░██║░░░██║╚██╔╝██║ ██║╚████║██╔══╝░░░░░██║░░░░░████╔═████║░██║░░██║██╔══██╗██╔═██╗░███████╗██████╔╝░░░██║░░░██║░╚═╝░██║ ██║░╚███║███████╗░░░██║░░░░░╚██╔╝░╚██╔╝░╚█████╔╝██║░░██║██║░╚██╗╚══════╝╚═════╝░░░░╚═╝░░░╚═╝░░░░░╚═╝ ╚═╝░░╚══╝╚══════╝░░░╚═╝░░░░░░╚═╝░░░╚═╝░░░╚════╝░╚═╝░░╚═╝╚═╝░░╚═╝This class handles the single-height LSTM network. The LSTM network comprises of an array of LSTM cell objects, an input matrix, and a target matrix. Forward propagation for the network, and backward network propagation is implemented here. This is seperate from the cellular-level forward and back propagation in the LSTM_CELL class.

Instance Method Summary collapse

Instance Method Details

#applyWeightChangeObject



158
159
160
161
162
# File 'lib/NETWORK.rb', line 158

def applyWeightChange()
	for i in 0...@lstm_nodes.column_count()
		@lstm_nodes[0,i].applyWeightChange()
	end
end

#backwardPropagateObject



140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
# File 'lib/NETWORK.rb', line 140

def backwardPropagate()
	i = 0
	delta_h_init = DFloat.zeros(@lstm_nodes[0, i].getHt().shape())
       delta_h_init = 2 * (@lstm_nodes[0, i].getYt() - @lstm_nodes[0, i].getHt())
	delta_c_init = DFloat.zeros(1, @sz)
	@lstm_nodes[0,i].backwardPropagation(delta_h_init, delta_c_init)

	i += 1
	while i <= @lstm_nodes.column_count()-1 do
		#puts "BP cell: " + i.to_s
		delta_h = DFloat.zeros(@lstm_nodes[0, i].getHt().shape())
       	delta_h = 2 * (@lstm_nodes[0, i].getYt() - @lstm_nodes[0, i].getHt())
		delta_h += @lstm_nodes[0, i-1].getBottomDeltaHt()
		delta_c = @lstm_nodes[0, i-1].getBottomDeltaCt()
		@lstm_nodes[0,i].backwardPropagation(delta_h, delta_c)
		i += 1
	end
end

#forwardPropagate(initialH = DFloat.zeros(1, @sz), initialC = DFloat.zeros(1, @sz)) ⇒ Object



108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
# File 'lib/NETWORK.rb', line 108

def forwardPropagate(initialH=DFloat.zeros(1, @sz), initialC=DFloat.zeros(1, @sz))
	@Output = DFloat.zeros(@Nodes, @sz)
	# Start from front of network and work forwards
	if initialH != nil && initialC != nil
		@lstm_nodes[0,0].setHprev(initialH)
		@lstm_nodes[0,0].setCprev(initialC)
	else 
		@lstm_nodes[0,0].setHprev(DFloat.zeros(1, @sz))
		@lstm_nodes[0,0].setCprev(DFloat.zeros(1, @sz))
	end
	
	node_input 	= DFloat[*[@Input[0, true].to_a]]
	node_target = DFloat[*[@Target[0, true].to_a]]

	@lstm_nodes[0,0].setXt(node_input)
	@lstm_nodes[0,0].setYt(node_target)
	@lstm_nodes[0,0].forwardPropagation()

	@Output[0, true] = @lstm_nodes[0,0].getHt()
	# Nodes 1 to end
	for i in 1...@lstm_nodes.column_count()
		# Indexing is working now
		node_input 	= DFloat[*[@Input[i, true].to_a]]
		node_target = DFloat[*[@Target[i, true].to_a]]
		@lstm_nodes[0,i].setHprev(@lstm_nodes[0,i-1].getHt())
		@lstm_nodes[0,i].setCprev(@lstm_nodes[0,i-1].getCt())
		@lstm_nodes[0,i].setXt(node_input)
		@lstm_nodes[0,i].setYt(node_target)
		@lstm_nodes[0,i].forwardPropagation()
		@Output[i, true] = @lstm_nodes[0,i].getHt()
	end
end

#getInput(mode = "word_mode") ⇒ Object



90
91
92
93
94
95
96
97
98
# File 'lib/NETWORK.rb', line 90

def getInput(mode="word_mode")
	if mode == "word_mode"
		return @dict.decodeArray(@Input)
	elsif mode == "char_mode"
		return @encoder.hotDecodeSentance(@encoder.nArrayToMatrix(@Input))
	elsif mode == "array_mode"
		return @Input
	end
end

#getLSTMNodesObject



53
54
55
# File 'lib/NETWORK.rb', line 53

def getLSTMNodes()
	return @lstm_nodes
end

#getOutput(mode = "word_mode") ⇒ Object



99
100
101
102
103
104
105
106
107
# File 'lib/NETWORK.rb', line 99

def getOutput(mode="word_mode")
	if mode == "word_mode"
		return @dict.decodeArrayByMaximum(@Output)
	elsif mode == "char_mode"
		return @encoder.hotDecodeSentance(@encoder.nArrayToMatrix(@Output))
	elsif mode == "array_mode"
		return @Output
	end
end

#getTarget(mode = "word_mode") ⇒ Object



81
82
83
84
85
86
87
88
89
# File 'lib/NETWORK.rb', line 81

def getTarget(mode="word_mode")
	if mode == "word_mode"
		return @dict.decodeArray(@Target)
	elsif mode == "char_mode"
		return @encoder.hotDecodeSentance(@encoder.nArrayToMatrix(@Target))
	elsif mode == "array_mode"
		return @Target
	end
end

#init(nodes, x_dim, alpha, terminal_output = nil) ⇒ Object



36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
# File 'lib/NETWORK.rb', line 36

def init(nodes, x_dim, alpha, terminal_output=nil)
	@dict = DICTIONARY. new
	@encoder = ENCODER. new
	@encoder.init()
	@sz = x_dim
	@@Alpha = alpha
	@Nodes = nodes
	### Creating LSTM network (matrix of nodes)
	@lstm_nodes = Matrix.build(1, @Nodes) { LSTM_CELL }
	for i in 0...@lstm_nodes.column_count()
		@lstm_nodes[0,i] = LSTM_CELL. new
		@lstm_nodes[0,i].init(@@Alpha, @sz, terminal_output)
	end
	## Input and target
	@Input = DFloat.zeros(nodes, @sz)
	@Target = DFloat.zeros(nodes, @sz)
end

#setDictionary(dictionary) ⇒ Object



56
57
58
# File 'lib/NETWORK.rb', line 56

def setDictionary(dictionary)
	@dict = dictionary
end

#setInput(input, mode = nil) ⇒ Object



70
71
72
73
74
75
76
77
78
79
80
# File 'lib/NETWORK.rb', line 70

def setInput(input, mode=nil)
	if mode == "encoded"
		@Input = input
	elsif input.instance_of? String
		@Input = @encoder.matrixToNArray(@encoder.hotEncodeSentance(input))
	elsif input.instance_of? Array
		@Input = @dict.encodeArray(input)
	else
		raise "Target input is not of type Array or String"
	end
end

#setTarget(target, mode = nil) ⇒ Object



59
60
61
62
63
64
65
66
67
68
69
# File 'lib/NETWORK.rb', line 59

def setTarget(target, mode=nil)
	if mode == "encoded"
		@Target = target
	elsif target.instance_of? String
		@Target = @encoder.matrixToNArray(@encoder.hotEncodeSentance(target))
	elsif target.instance_of? Array
		@Target = @dict.encodeArray(target)
	else
		raise "Target input is not of type Array or String"
	end
end