Class: DNN::Layers::LSTM

Inherits:
RNN show all
Defined in:
lib/dnn/core/rnn_layers.rb

Instance Attribute Summary

Attributes inherited from RNN

#num_nodes, #recurrent_weight_initializer, #recurrent_weight_regularizer, #return_sequences, #stateful

Attributes inherited from Connection

#bias_initializer, #bias_regularizer, #weight_initializer, #weight_regularizer

Attributes inherited from HasParamLayer

#params, #trainable

Attributes inherited from Layer

#input_shape, #learning_phase

Class Method Summary collapse

Instance Method Summary collapse

Methods inherited from RNN

#output_shape, #regularizers, #to_hash

Methods inherited from Connection

#regularizers, #to_hash, #use_bias

Methods inherited from Layer

#built?, #output_shape, #to_hash

Constructor Details

#initialize(num_nodes, stateful: false, return_sequences: true, weight_initializer: RandomNormal.new, recurrent_weight_initializer: RandomNormal.new, bias_initializer: Zeros.new, weight_regularizer: nil, recurrent_weight_regularizer: nil, bias_regularizer: nil, use_bias: true) ⇒ LSTM

Returns a new instance of LSTM.



281
282
283
284
285
286
287
288
289
290
291
292
293
 
# File 'lib/dnn/core/rnn_layers.rb', line 281

def initialize(num_nodes,
               stateful: false,
               return_sequences: true,
               weight_initializer: RandomNormal.new,
               recurrent_weight_initializer: RandomNormal.new,
               bias_initializer: Zeros.new,
               weight_regularizer: nil,
               recurrent_weight_regularizer: nil,
               bias_regularizer: nil,
               use_bias: true)
  super
  @cell = @params[:cell] = Param.new
end

Class Method Details

.from_hash(hash) ⇒ Object 



267
268
269
270
271
272
273
274
275
276
277
278
279
 
# File 'lib/dnn/core/rnn_layers.rb', line 267

def self.from_hash(hash)
  lstm = self.new(hash[:num_nodes],
                  stateful: hash[:stateful],
                  return_sequences: hash[:return_sequences],
                  weight_initializer: Utils.from_hash(hash[:weight_initializer]),
                  recurrent_weight_initializer: Utils.from_hash(hash[:recurrent_weight_initializer]),
                  bias_initializer: Utils.from_hash(hash[:bias_initializer]),
                  weight_regularizer: Utils.from_hash(hash[:weight_regularizer]),
                  recurrent_weight_regularizer: Utils.from_hash(hash[:recurrent_weight_regularizer]),
                  bias_regularizer: Utils.from_hash(hash[:bias_regularizer]),
                  use_bias: hash[:use_bias])
  lstm
end

Instance Method Details

#backward(dh2s) ⇒ Object 



329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
 
# File 'lib/dnn/core/rnn_layers.rb', line 329

def backward(dh2s)
  unless @return_sequences
    dh = dh2s
    dh2s = Xumo::SFloat.zeros(dh.shape[0], @time_length, dh.shape[1])
    dh2s[true, -1, false] = dh
  end
  dxs = Xumo::SFloat.zeros(@xs_shape)
  dh = 0
  dc = 0
  (0...dh2s.shape[1]).to_a.reverse.each do |t|
    dh2 = dh2s[true, t, false]
    dx, dh, dc = @layers[t].backward(dh2 + dh, dc)
    dxs[true, t, false] = dx
  end
  dxs
end

#build(input_shape) ⇒ Object 



295
296
297
298
299
300
301
302
303
304
305
 
# File 'lib/dnn/core/rnn_layers.rb', line 295

def build(input_shape)
  super
  num_prev_nodes = input_shape[1]
  @weight.data = Xumo::SFloat.new(num_prev_nodes, @num_nodes * 4)
  @recurrent_weight.data = Xumo::SFloat.new(@num_nodes, @num_nodes * 4)
  @bias.data = Xumo::SFloat.new(@num_nodes * 4) if @bias
  init_weight_and_bias
  @time_length.times do |t|
    @layers << LSTM_Dense.new(@weight, @recurrent_weight, @bias)
  end
end

#forward(xs) ⇒ Object 



307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
 
# File 'lib/dnn/core/rnn_layers.rb', line 307

def forward(xs)
  @xs_shape = xs.shape
  hs = Xumo::SFloat.zeros(xs.shape[0], @time_length, @num_nodes)
  h = nil
  c = nil
  if @stateful
    h = @hidden.data if @hidden.data
    c = @cell.data if @cell.data
  end
  h ||= Xumo::SFloat.zeros(xs.shape[0], @num_nodes)
  c ||= Xumo::SFloat.zeros(xs.shape[0], @num_nodes)
  xs.shape[1].times do |t|
    x = xs[true, t, false]
    @layers[t].trainable = @trainable
    h, c = @layers[t].forward(x, h, c)
    hs[true, t, false] = h
  end
  @hidden.data = h
  @cell.data = c
  @return_sequences ? hs : h
end

#reset_state ⇒ Object 



346
347
348
349
 
# File 'lib/dnn/core/rnn_layers.rb', line 346

def reset_state
  super()
  @cell.data = @cell.data.fill(0) if @cell.data
end