vw(1)

VW options:

-h [ --help ]

Look here: http://hunch.net/~vw/ and click on Tutorial.

--active_learning

active learning mode

--active_simulation

active learning simulation mode

--active_mellowness arg

active learning mellowness parameter c_0. Default 8

--binary

report loss as binary classification on -1,1

--autolink arg

create link function with polynomial d

--sgd

use regular stochastic gradient descent update.

--adaptive

use adaptive, individual learning rates.

--invariant

use safe/importance aware updates.

--normalized

use per feature normalized updates

--exact_adaptive_norm

use current default invariant normalized adaptive update rule

-a [ --audit ]

print weights of features

-b [ --bit_precision ] arg

number of bits in the feature table

--bfgs

use bfgs optimization

-c [ --cache ]

Use a cache. The default is <data>.cache

--cache_file arg

The location(s) of cache_file.

--compressed

use gzip format whenever possible. If a cache file is being created, this option creates a compressed cache file. A mixture of raw-text & compressed inputs are supported with autodetection.

--no_stdin

do not default to reading from stdin

--conjugate_gradient

use conjugate gradient based optimization

--csoaa arg

Use one-against-all multiclass learning with <k> costs

--wap arg

Use weighted all-pairs multiclass learning with <k> costs

--csoaa_ldf arg

Use one-against-all multiclass learning with label dependent features. Specify singleline or multiline.

--wap_ldf arg

Use weighted all-pairs multiclass learning with label dependent features.

Specify singleline or multiline.

--cb arg

Use contextual bandit learning with <k> costs

--l1 arg

l_1 lambda

--l2 arg

l_2 lambda

-d [ --data ] arg

Example Set

--daemon

persistent daemon mode on port 26542

--num_children arg

number of children for persistent daemon mode

--pid_file arg

Write pid file in persistent daemon mode

--decay_learning_rate arg

Set Decay factor for learning_rate between passes

--input_feature_regularizer arg

Per feature regularization input file

-f [ --final_regressor ] arg

Final regressor

--readable_model arg

Output human-readable final regressor

--hash arg

how to hash the features. Available options: strings, all

--hessian_on

use second derivative in line search

--version

Version information

--ignore arg

ignore namespaces beginning with character <arg>

--keep arg

keep namespaces beginning with character <arg>

-k [ --kill_cache ]

do not reuse existing cache: create a new one always

--initial_weight arg

Set all weights to an initial value of 1.

-i [ --initial_regressor ] arg

Initial regressor(s)

--initial_pass_length arg

initial number of examples per pass

--initial_t arg

initial t value

--lda arg

Run lda with <int> topics

--span_server arg

Location of server for setting up spanning tree

--min_prediction arg

Smallest prediction to output

--max_prediction arg

Largest prediction to output

--mem arg

memory in bfgs

--nn arg

Use sigmoidal feedforward network with <k> hidden units

--noconstant

Don't add a constant feature

--noop

do no learning

--oaa arg

Use one-against-all multiclass learning with <k> labels

--ect arg

Use error correcting tournament with <k> labels

--output_feature_regularizer_binary arg

Per feature regularization output file

--output_feature_regularizer_text arg Per feature regularization output file,

in text

--port arg

port to listen on

--power_t arg

t power value

-l [ --learning_rate ] arg

Set Learning Rate

--passes arg

Number of Training Passes

--termination arg

Termination threshold

-p [ --predictions ] arg

File to output predictions to

-q [ --quadratic ] arg

Create and use quadratic features

--cubic arg

Create and use cubic features

--quiet

Don't output diagnostics

--rank arg

rank for matrix factorization.

--random_weights arg

make initial weights random

--random_seed arg

seed random number generator

-r [ --raw_predictions ] arg

File to output unnormalized predictions to

--ring_size arg

size of example ring

--examples arg

number of examples to parse

--save_per_pass

Save the model after every pass over data

--save_resume

save extra state so learning can be resumed later with new data

--sendto arg

send examples to <host>

--searn arg

use searn, argument=maximum action id

--searnimp arg

use searn, argument=maximum action id or 0 for LDF

-t [ --testonly ]

Ignore label information and just test

--loss_function arg (=squared)

Specify the loss function to be used, uses squared by default. Currently available ones are squared, classic, hinge, logistic and quantile.

--quantile_tau arg (=0.5)

Parameter \tau associated with Quantile loss. Defaults to 0.5

--unique_id arg

unique id used for cluster parallel jobs

--total arg

total number of nodes used in cluster parallel job

--node arg

node number in cluster parallel job

--sort_features

turn this on to disregard order in which features have been defined. This will lead to smaller cache sizes

--ngram arg

Generate N grams

--skips arg

Generate skips in N grams. This in conjunction with the ngram tag can be used to generate generalized n-skip-k-gram.