Functions — github.com/adamsolomou/SC-DNN

hub / github.com/adamsolomou/SC-DNN / functions

Functions46 in github.com/adamsolomou/SC-DNN

Linear gain with saturation. The implementation currently supports only a linear gain of 2, 4 and 8. Parameters ---------- x:

src/inference/function_blocks.py:614

↓ 6 callersFunctionsng

Bipolar stochastic number generator Parameters ---------- x: Floating-point input value no_stoch_samples: Bit-stream length (i

src/inference/function_blocks.py:24

↓ 4 callersFunctionget_value

Estimate the value of a stochastic bit-stream Parameters ---------- x: 1D numpy array holding the input bit-stream Returns

src/inference/function_blocks.py:92

↓ 4 callersFunctioninvert

(x)

src/inference/function_blocks.py:173

↓ 4 callersFunctionmultiply

Stochastic XNOR multiplier Parameters ---------- x: Stochastic bit-stream w: Floating point weight Returns -------

src/inference/function_blocks.py:146

↓ 4 callersFunctionregularizer

(x)

src/training/sc_train_creg.py:11

↓ 3 callersFunctionadd

Scaled addition of two stochastic bit-streams Parameters ---------- x,y: 1D numpy arrays holding bit-streams to be added s_in:

src/inference/function_blocks.py:178

↓ 2 callersFunctionSmax

Stochastic max unit Parameters ---------- a,b: Input bit-streams N: Integer specifying the number of states Returns -

src/inference/function_blocks.py:807

↓ 2 callersFunctionStanh

Stochastic hyperbolic tangent Parameters ---------- x: Input stochastic bit-stream N: Integer specifying the number of states

src/inference/function_blocks.py:568

↓ 2 callersFunctionnext_int_power_of_2

Vectorized form of scalar_next_int_power_of_2

src/inference/scaling_ops.py:10

↓ 2 callersFunctionsc_add

Parameters ---------- x: Input data 2D tensor w: Weight 2D tensor s_in: 1D tensor with scaling factors of the input activations

src/training/sc_train_creg.py:114

↓ 2 callersFunctionsc_add

Parameters ---------- x: Input data 2D tensor w: Weight 2D tensor s_in: 1D tensor with scaling factors of the input activations

src/training/sc_train_l2reg.py:114

↓ 2 callersFunctionsc_matmul

Parameters ---------- x: Input data 2D tensor w: Weight 2D tensor s_in: 1D tensor with scaling factors of the input activations

src/training/sc_train_creg.py:68

↓ 2 callersFunctionsc_matmul

Parameters ---------- x: Input data 2D tensor w: Weight 2D tensor s_in: 1D tensor with scaling factors of the input activations

src/training/sc_train_l2reg.py:68

↓ 1 callersFunctionSmin

Stochastic min unit Parameters ---------- a,b: Input bit-streams N: Integer specifying the number of states Returns -

src/inference/function_blocks.py:843

↓ 1 callersFunctionSrelu

Stochastic approximation of the ReLU Parameters ---------- x: Input stochastic bit-stream N: Integer specifying the number of s

src/inference/function_blocks.py:879

↓ 1 callersFunctiondec_dot

Stochastic inner product with decomposition. The implementation assumes that the input bit-streams have a common scaling parameter. Par

src/inference/function_blocks.py:364

↓ 1 callersFunctiondo_it

()

src/inference/function_blocks.py:527

↓ 1 callersFunctiondot

Scaled inner product in stochastic computing. The implementation assumes that the input bit-streams have a common scaling parameter. Pa

src/inference/function_blocks.py:291

↓ 1 callersFunctionmain

()

src/inference/network_II/sc_inference.py:12

↓ 1 callersFunctionmain

()

src/inference/network_I/sc_inference.py:12

↓ 1 callersFunctionmat_sng

Vectorised form of sng Parameters ---------- x: 2D numpy array of floating-point input values no_stoch_samples: Bit-stream lengt

src/inference/function_blocks.py:68

↓ 1 callersFunctionneural_net

(x)

src/inference/network_II/fp_inference.py:42

↓ 1 callersFunctionneural_net

(x)

src/inference/network_I/fp_inference.py:42

↓ 1 callersFunctionnext_int_power_of_2

Vectorized form of scalar_next_int_power_of_2

src/inference/function_blocks.py:17

↓ 1 callersFunctionpopcount

(x)

src/inference/function_blocks.py:547

↓ 1 callersFunctionv_add_scale

Vectorized form of add_scale() function

src/inference/scaling_ops.py:69

↓ 1 callersFunctionvec_sng

Vectorised form of sng Parameters ---------- x: 1D numpy array of floating-point input values no_stoch_samples: Bit-stream lengt

src/inference/function_blocks.py:46

FunctionSabs

Stochastic approximation of the absolute value function Parameters ---------- x: Stochastic bit-stream N: Integer specifying th

src/inference/function_blocks.py:755

FunctionSexp

Stochastic exponentiation function Note: G should be such that G<<N Parameters ---------- x: Stochastic bit-stream G: Posi

src/inference/function_blocks.py:706

Functionadd_scale

Parameters ---------- s1,s2: Input scaling parameters Returns ------- The output scaling of a two input scaled added

src/inference/scaling_ops.py:57

Functionhelper

(row_idx,queue)

src/inference/function_blocks.py:496

Functionmat_sc_value

Vectorised version of the get_value function Parameters ---------- x: 3D numpy array with stochastic bit-streams

src/inference/function_blocks.py:126

Functionnon_lin_sat

Approximates a non-linear block

src/inference/function_blocks.py:550

Functionrelu

Returns the output scaling of a ReLU

src/inference/scaling_ops.py:114

Functionsc_matmul

Stochastic matrix multiplication of the form X*W. The implementation assumes that the input bit-streams have a common scaling parameter.

src/inference/function_blocks.py:456

Functionsc_matmul_scaling

The function computes the output scalings of a stochastic computing based matrix multiplication of the form X*W, where X is a 3D numpy

src/inference/scaling_ops.py:16

Functionsc_matvec_add

Matrix to vector addition of the form X + b. The vector b is broaccast along the rows of X. The implementation assumes that the inputs have a

src/inference/function_blocks.py:242

Functionsc_matvec_add_scaling

The function computes the output scalings of a stochastic computing based matrix addition of the form X + b, where X is a 3D numpy array

src/inference/scaling_ops.py:75

Functionscalar_next_int_power_of_2

Return the next integer power of 2 of x

src/inference/function_blocks.py:10

Functionscalar_next_int_power_of_2

Return the next integer power of 2

src/inference/scaling_ops.py:4

Functionsquare

(x)

src/inference/function_blocks.py:165

FunctionvSmax

Compute the max value from the list

src/inference/function_blocks.py:834

FunctionvSmin

Compute the min value from the list

src/inference/function_blocks.py:870

Functionvec_Srelu

Vectorised form of Srelu

src/inference/function_blocks.py:895

Functionvec_sc_value

Vectorised version of the get_value function Parameters ---------- x: 2D numpy array with stochastic bit-streams

src/inference/function_blocks.py:108