Package model_descriptions :: Module linear_gradients

[frames] | no frames]

Module linear_gradients

source code

This module contains functions for linear gradients

Functions

[hide private]

tuple

ell2_ball_constraint_avg(w, w_hat, steps, x, y, mu, B, stepsize)
Calculates and returns ell2 ball constraint avg

source code

double

hinge_loss(w, x, y)
Calculates and returns hinge loss

source code

double

sparse_hinge_loss(w, indexes, vecs, y)
Calculates hinge lss for sparse vectors and returns hinge loss value

source code

double

mse_loss(w, x, y)
Calculates and returns mse loss

source code

double

logit_loss_ss(w, index, vecs, y)
Calculates logit loss for sparse vectors and returns the value

source code

test()
Tests ell2 ball constraint and hinge loss functions

source code

lazy_shrink(w, current_time, timestamps, indexes, u)
Lazy shrink function

source code

incremental_average_sparse(w_hat, w, steps, indexes)
Incremental average sparse function

source code

l1_prox_lazy(w, what, timestamps, current_time, steps, stepsize, mu, B, indexes, vecs, y)
l1 prox lazy function

source code

svm_l1_prox_lazy(w, what, timestamps, current_time, steps, stepsize, mu, B, indexes, vecs, y)
svm l1 prox lazy function

source code

svm_l2_prox_lazy(w, what, wscale, steps, stepsize, mu, B, indexes, vecs, y)
svm l2 prox lazy function

source code

Function Details

[hide private]

ell2_ball_constraint_avg(w, w_hat, steps, x, y, mu, B, stepsize)

source code

Calculates and returns ell2 ball constraint avg

Parameters:

w (vector) - model vector
w_hat (vector) - w_hat vector
steps (number) - number of steps
x (vector) - feature vector of the example
y (number) - label of the example
mu (double) - mu
B (double) - B
stepsize (double) - step size

Returns: tuple

returns w, w_hat, # of steps, mu and B

hinge_loss(w, x, y)

source code

Calculates and returns hinge loss

Parameters:

w (vector) - model vector
x (vector) - feature vector of the example
y (number) - label of the example

Returns: double

returns the hinge loss

sparse_hinge_loss(w, indexes, vecs, y)

source code

Calculates hinge lss for sparse vectors and returns hinge loss value

Parameters:

w (vector) - model vector
indexes (vector) - indexes of the feature vector of the example
vecs (vector) - vector values of the feature vector of the example
y (number) - label of the example

Returns: double

returns the hinge loss

mse_loss(w, x, y)

source code

Calculates and returns mse loss

Parameters:

w (vector) - model vector
x (vector) - feature vector of the example
y (number) - label of the example

Returns: double

returns the mse loss

logit_loss_ss(w, index, vecs, y)

source code

Calculates logit loss for sparse vectors and returns the value

Parameters:

w (vector) - model vector
index (vector) - indexes of the feature vector of the example
vecs (vector) - vector values of the feature vector of the example
y (number) - label of the example

Returns: double

returns the logit loss

lazy_shrink(w, current_time, timestamps, indexes, u)

source code

Lazy shrink function

Parameters:

w (vector) - model vector
current_time (double) - current time
timestamps (vector) - time stamps vector
indexes (vector) - indexes of the example
u (double) - u

incremental_average_sparse(w_hat, w, steps, indexes)

source code

Incremental average sparse function

Parameters:

w_hat (vector) - w hat vector
w (vector) - model vector
steps (double) - steps of the incremental stage
indexes (vector) - indexes

l1_prox_lazy(w, what, timestamps, current_time, steps, stepsize, mu, B, indexes, vecs, y)

source code

l1 prox lazy function

Parameters:

w (vector) - model vector
what (vector) - w hat
timestamps (vector) - time stamps vector
current_time (double) - current time
steps (double) - steps of the incremental stage
stepsize (number) - step size
mu (double) - mu
B (double) - B
indexes (vector) - indexes vector of the example's feature vector
vecs (vector) - vector values of the example's feature vector
y (number) - label of the example

svm_l1_prox_lazy(w, what, timestamps, current_time, steps, stepsize, mu, B, indexes, vecs, y)

source code

svm l1 prox lazy function

Parameters:

w (vector) - model vector
what (vector) - w hat
timestamps (vector) - time stamps vector
current_time (double) - current time
steps (double) - steps of the incremental stage
stepsize (number) - step size
mu (double) - mu
B (double) - B
indexes (vector) - indexes vector of the example's feature vector
vecs (vector) - vector values of the example's feature vector
y (number) - label of the example

svm_l2_prox_lazy(w, what, wscale, steps, stepsize, mu, B, indexes, vecs, y)

source code

svm l2 prox lazy function

Parameters:

w (vector) - model vector
what (vector) - w hat
wscale (double) - used to scale model vector(w)
steps (double) - steps of the incremental stage
stepsize (number) - step size
mu (double) - mu
B (double) - B
indexes (vector) - indexes vector of the example's feature vector
vecs (vector) - vector values of the example's feature vector
y (number) - label of the example