RNN

Input

• x:
  • Required: yes
  • Type: float16
  • Description: time change input, (T × N × ...)
  • Restriction: Must be within 256.
• cont:
  • Required: yes
  • Type: float16
  • Description: sequence continuity flag (T × N)
  • Restriction: Must be within 256.
• x_static (optional):
  • Required: no
  • Type: float16
  • Description: static data with time unchanged (N × ...)
  • Restriction: none
• h_0 (optional):
  • Required: no
  • Type: float16, float32
  • Description: initial hidden state (1×N×num_output)
  • Restriction: none
• w_xh:
  • Required: yes
  • Type: float16
  • Description: xH weight, with shape [input_size, num_output]
  • Restriction: none
• bias_h:
  • Required: yes
  • Type: float16, float32
  • Description: bias, with shape [num_output,]
  • Restriction: none
• w_sh:
  • Required: no
  • Type: float16
  • Description: sH weight, with shape [input_size, num_output]
  • Restriction: none
• w_hh:
  • Required: yes
  • Type: float16
  • Description: hH weight, with shape [num_output, num_output]
  • Restriction: none
• w_ho:
  • Required: yes
  • Type: float16
  • Description: hO weight, with shape [num_output, num_output]
  • Restriction: none
• bias_o:
  • Required: yes
  • Type: float16, float32
  • Description: bias, with shape [num_output,]
  • Restriction: none

Attribute

• expose_hidden:
  • Required: no
  • Type: bool
  • Description: whether to expose the hidden status (default = false)
  • Restriction: none

Output

• o:
  • Required: yes
  • Type: float16, float32
  • Description: output tensor (T x N x num_output)
  • Restriction: none

• h_t:
  • Required: yes
  • Type: float16, float32
  • Description: hidden status of the last time step (1 * N * num_output)
  • Restriction: none