LSTM - 1 vs 7#

Next section compares an older to a newer version of the same operator after both definition are converted into markdown text. Green means an addition to the newer version, red means a deletion. Anything else is unchanged.

Files changed (1) hide show
  1. LSTM1 → LSTM7 +8 -7
LSTM1 → LSTM7 RENAMED
@@ -1 +1 @@
1
1
  Computes an one-layer LSTM. This operator is usually supported via some
2
2
  custom implementation such as CuDNN.
3
3
  Notations:
4
4
  X - input tensor
5
5
  i - input gate
6
6
  o - output gate
7
7
  f - forget gate
8
8
  c - cell gate
9
9
  t - time step (t-1 means previous time step)
10
10
  W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
11
11
  R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
12
12
  Wb[iofc] - W bias vectors for input, output, forget, and cell gates
13
13
  Rb[iofc] - R bias vectors for input, output, forget, and cell gates
14
14
  P[iof] - P peephole weight vector for input, output, and forget gates
15
15
  WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
16
16
  RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
17
17
  WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
18
18
  RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
19
19
  PB[iof] - P peephole weight vector for backward input, output, and forget gates
20
20
  H - Hidden state
21
21
  num_directions - 2 if direction == bidirectional else 1
22
22
  Activation functions:
23
23
  Relu(x) - max(0, x)
24
24
  Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
25
25
  Sigmoid(x) - 1/(1 + e^{-x})
26
26
  (NOTE: Below are optional)
27
27
  Affine(x) - alpha*x + beta
28
28
  LeakyRelu(x) - x if x >= 0 else alpha * x
29
29
  ThresholdedRelu(x) - x if x >= alpha else 0
30
30
  ScaledTanh(x) - alpha*Tanh(beta*x)
31
31
  HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
32
32
  Elu(x) - x if x >= 0 else alpha*(e^x - 1)
33
33
  Softsign(x) - x/(1 + |x|)
34
34
  Softplus(x) - log(1 + e^x)
35
35
  Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
36
- - it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
36
+ - it = f(Xt*(Wi^T) + Ht-1*Ri + Pi (.) Ct-1 + Wbi + Rbi)
37
- - ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
37
+ - ft = f(Xt*(Wf^T) + Ht-1*Rf + Pf (.) Ct-1 + Wbf + Rbf)
38
- - ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
38
+ - ct = g(Xt*(Wc^T) + Ht-1*Rc + Wbc + Rbc)
39
39
  - Ct = ft (.) Ct-1 + it (.) ct
40
- - ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
40
+ - ot = f(Xt*(Wo^T) + Ht-1*Ro + Po (.) Ct + Wbo + Rbo)
41
41
  - Ht = ot (.) h(Ct)
42
- This operator has **optional** inputs/outputs. See ONNX <https://github.com/onnx/onnx/blob/master/docs/IR.md>_ for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
43
42
  **Attributes**
44
43
  * **activation_alpha**:
45
44
  Optional scaling values used by some activation functions. The
46
45
  values are consumed in the order of activation functions, for
47
46
  example (f, g, h) in LSTM. Default values are the same as of
48
47
  corresponding ONNX operators.For example with LeakyRelu, the default
49
48
  alpha is 0.01.
50
49
  * **activation_beta**:
51
50
  Optional scaling values used by some activation functions. The
52
51
  values are consumed in the order of activation functions, for
53
52
  example (f, g, h) in LSTM. Default values are the same as of
54
53
  corresponding ONNX operators.
55
54
  * **activations**:
56
55
  A list of 3 (or 6 if bidirectional) activation functions for input,
57
56
  output, forget, cell, and hidden. The activation functions must be
58
57
  one of the activation functions specified above. Optional: See the
59
58
  equations for default if not specified.
60
59
  * **clip**:
61
60
  Cell clip threshold. Clipping bounds the elements of a tensor in the
62
61
  range of [-threshold, +threshold] and is applied to the input of
63
62
  activations. No clip if not specified.
64
63
  * **direction**:
65
64
  Specify if the RNN is forward, reverse, or bidirectional. Must be
66
65
  one of forward (default), reverse, or bidirectional.
67
66
  * **hidden_size**:
68
67
  Number of neurons in the hidden layer
69
68
  * **input_forget**:
70
- Couple the input and forget gates if 1.
69
+ Couple the input and forget gates if 1, default 0.
70
+ * **output_sequence**:
71
+ The sequence output for the hidden is optional if 0. Default 0.
71
72
  **Inputs**
72
73
  Between 3 and 8 inputs.
73
74
  * **X** (heterogeneous) - **T**:
74
75
  The input sequences packed (and potentially padded) into one 3-D
75
76
  tensor with the shape of [seq_length, batch_size, input_size].
76
77
  * **W** (heterogeneous) - **T**:
77
78
  The weight tensor for the gates. Concatenation of W[iofc] and
78
79
  WB[iofc] (if bidirectional) along dimension 0. The tensor has
79
80
  shape [num_directions, 4*hidden_size, input_size].
80
81
  * **R** (heterogeneous) - **T**:
81
82
  The recurrence weight tensor. Concatenation of R[iofc] and
82
83
  RB[iofc] (if bidirectional) along dimension 0. This tensor has
83
84
  shape [num_directions, 4*hidden_size, hidden_size].
84
85
  * **B** (optional, heterogeneous) - **T**:
85
86
  The bias tensor for input gate. Concatenation of [Wb[iofc],
86
87
  Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along
87
88
  dimension 0. This tensor has shape [num_directions,
88
89
  8*hidden_size]. Optional: If not specified - assumed to be 0.
89
90
  * **sequence_lens** (optional, heterogeneous) - **T1**:
90
91
  Optional tensor specifying lengths of the sequences in a batch. If
91
92
  not specified - assumed all sequences in the batch to have length
92
93
  seq_length. It has shape [batch_size].
93
94
  * **initial_h** (optional, heterogeneous) - **T**:
94
95
  Optional initial value of the hidden. If not specified - assumed to
95
96
  be 0. It has shape [num_directions, batch_size, hidden_size].
96
97
  * **initial_c** (optional, heterogeneous) - **T**:
97
98
  Optional initial value of the cell. If not specified - assumed to be
98
99
  0. It has shape [num_directions, batch_size, hidden_size].
99
100
  * **P** (optional, heterogeneous) - **T**:
100
101
  The weight tensor for peepholes. Concatenation of P[iof] and
101
102
  PB[iof] (if bidirectional) along dimension 0. It has shape
102
103
  [num_directions, 3*hidde_size]. Optional: If not specified -
103
104
  assumed to be 0.
104
105
  **Outputs**
105
106
  Between 0 and 3 outputs.
106
107
  * **Y** (optional, heterogeneous) - **T**:
107
108
  A tensor that concats all the intermediate output values of the
108
109
  hidden. It has shape [seq_length, num_directions, batch_size,
109
- hidden_size].
110
+ hidden_size]. It is optional if output_sequence is 0.
110
111
  * **Y_h** (optional, heterogeneous) - **T**:
111
112
  The last output value of the hidden. It has shape [num_directions,
112
113
  batch_size, hidden_size].
113
114
  * **Y_c** (optional, heterogeneous) - **T**:
114
115
  The last output value of the cell. It has shape [num_directions,
115
116
  batch_size, hidden_size].
116
117
  **Type Constraints**
117
118
  * **T** in (
118
119
  tensor(double),
119
120
  tensor(float),
120
121
  tensor(float16)
121
122
  ):
122
123
  Constrain input and output types to float tensors.
123
124
  * **T1** in (
124
125
  tensor(int32)
125
126
  ):
126
127
  Constrain seq_lens to integer tensor.