LSTM - 1 vs 7

Files changed (1) hide show

1. LSTM1 → LSTM7 +7 -8

LSTM1 → LSTM7 RENAMED Viewed

	@@ -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 + Pi (.) Ct-1 + Wbi + Rbi)
36	+ - it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
37	- - ft = f(Xt*(Wf^T) + Ht-1*Rf + Pf (.) Ct-1 + Wbf + Rbf)
37	+ - ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
38	- - ct = g(Xt*(Wc^T) + Ht-1*Rc + Wbc + Rbc)
38	+ - ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
39 39	- Ct = ft (.) Ct-1 + it (.) ct
40	- - ot = f(Xt*(Wo^T) + Ht-1*Ro + Po (.) Ct + Wbo + Rbo)
40	+ - ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + 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.
42 43	**Attributes**
43 44	* **activation_alpha**:
44 45	Optional scaling values used by some activation functions. The
45 46	values are consumed in the order of activation functions, for
46 47	example (f, g, h) in LSTM. Default values are the same as of
47 48	corresponding ONNX operators.For example with LeakyRelu, the default
48 49	alpha is 0.01.
49 50	* **activation_beta**:
50 51	Optional scaling values used by some activation functions. The
51 52	values are consumed in the order of activation functions, for
52 53	example (f, g, h) in LSTM. Default values are the same as of
53 54	corresponding ONNX operators.
54 55	* **activations**:
55 56	A list of 3 (or 6 if bidirectional) activation functions for input,
56 57	output, forget, cell, and hidden. The activation functions must be
57 58	one of the activation functions specified above. Optional: See the
58 59	equations for default if not specified.
59 60	* **clip**:
60 61	Cell clip threshold. Clipping bounds the elements of a tensor in the
61 62	range of [-threshold, +threshold] and is applied to the input of
62 63	activations. No clip if not specified.
63 64	* **direction**:
64 65	Specify if the RNN is forward, reverse, or bidirectional. Must be
65 66	one of forward (default), reverse, or bidirectional.
66 67	* **hidden_size**:
67 68	Number of neurons in the hidden layer
68 69	* **input_forget**:
69	- Couple the input and forget gates if 1, default 0.
70	+ Couple the input and forget gates if 1.
70	- * **output_sequence**:
71	- The sequence output for the hidden is optional if 0. Default 0.
72 71	**Inputs**
73 72	Between 3 and 8 inputs.
74 73	* **X** (heterogeneous) - **T**:
75 74	The input sequences packed (and potentially padded) into one 3-D
76 75	tensor with the shape of [seq_length, batch_size, input_size].
77 76	* **W** (heterogeneous) - **T**:
78 77	The weight tensor for the gates. Concatenation of W[iofc] and
79 78	WB[iofc] (if bidirectional) along dimension 0. The tensor has
80 79	shape [num_directions, 4*hidden_size, input_size].
81 80	* **R** (heterogeneous) - **T**:
82 81	The recurrence weight tensor. Concatenation of R[iofc] and
83 82	RB[iofc] (if bidirectional) along dimension 0. This tensor has
84 83	shape [num_directions, 4*hidden_size, hidden_size].
85 84	* **B** (optional, heterogeneous) - **T**:
86 85	The bias tensor for input gate. Concatenation of [Wb[iofc],
87 86	Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along
88 87	dimension 0. This tensor has shape [num_directions,
89 88	8*hidden_size]. Optional: If not specified - assumed to be 0.
90 89	* **sequence_lens** (optional, heterogeneous) - **T1**:
91 90	Optional tensor specifying lengths of the sequences in a batch. If
92 91	not specified - assumed all sequences in the batch to have length
93 92	seq_length. It has shape [batch_size].
94 93	* **initial_h** (optional, heterogeneous) - **T**:
95 94	Optional initial value of the hidden. If not specified - assumed to
96 95	be 0. It has shape [num_directions, batch_size, hidden_size].
97 96	* **initial_c** (optional, heterogeneous) - **T**:
98 97	Optional initial value of the cell. If not specified - assumed to be
99 98	0. It has shape [num_directions, batch_size, hidden_size].
100 99	* **P** (optional, heterogeneous) - **T**:
101 100	The weight tensor for peepholes. Concatenation of P[iof] and
102 101	PB[iof] (if bidirectional) along dimension 0. It has shape
103 102	[num_directions, 3*hidde_size]. Optional: If not specified -
104 103	assumed to be 0.
105 104	**Outputs**
106 105	Between 0 and 3 outputs.
107 106	* **Y** (optional, heterogeneous) - **T**:
108 107	A tensor that concats all the intermediate output values of the
109 108	hidden. It has shape [seq_length, num_directions, batch_size,
110	- hidden_size]. It is optional if output_sequence is 0.
109	+ hidden_size].
111 110	* **Y_h** (optional, heterogeneous) - **T**:
112 111	The last output value of the hidden. It has shape [num_directions,
113 112	batch_size, hidden_size].
114 113	* **Y_c** (optional, heterogeneous) - **T**:
115 114	The last output value of the cell. It has shape [num_directions,
116 115	batch_size, hidden_size].
117 116	**Type Constraints**
118 117	* **T** in (
119 118	tensor(double),
120 119	tensor(float),
121 120	tensor(float16)
122 121	):
123 122	Constrain input and output types to float tensors.
124 123	* **T1** in (
125 124	tensor(int32)
126 125	):
127 126	Constrain seq_lens to integer tensor.