-
Notifications
You must be signed in to change notification settings - Fork 1.9k
Controlling VW's output
VW can produce several logical streams of output:
This describes the run. Configuration, progressive results and then final results.
This can be turned off with --driver_output_off
, or with --quiet
. By default, this goes to stderr
but --driver_output_stream [stdout|stderr]
can be used to select stdout
or stderr
.
only testing
predictions = 0001.predict
Num weight bits = 18
learning rate = 10
initial_t = 1
power_t = 0.5
using no cache
Reading datafile = train-sets/0001.dat
num sources = 1
Enabled reductions: gd, scorer-identity, count_label
Input label = simple
Output pred = scalar
average since example example current current current
loss last counter weight label predict features
0.000000 0.000000 1 1.0 1.0000 1.0000 290
0.000000 0.000000 2 2.0 0.0000 0.0000 608
0.000000 0.000000 4 4.0 0.0000 0.0000 794
0.000000 0.000000 8 8.0 0.0000 0.0000 860
0.000000 0.000000 16 16.0 1.0000 1.0000 128
0.000000 0.000000 32 32.0 0.0000 0.0000 176
0.000000 0.000000 64 64.0 0.0000 0.0000 350
0.000000 0.000000 128 128.0 1.0000 1.0000 620
finished run
number of examples = 200
weighted example sum = 200.000000
weighted label sum = 91.000000
average loss = 0.000000
best constant = 0.455000
best constant's loss = 0.247975
total feature number = 89692
VW produces log messages during operation. --log_level [info|warning|error|critical|off]
can be used to filter which levels to show. --quiet
implies --log_level off
and takes higher priority.
The destination stream can be selected with --log_output_stream [stdout|stderr|compat]
. In the past, VW produced log like messages to a mix of both stdout
and stderr
, compat
maintains this old behavior is the current default. stdout
or stderr
should be chosen as compat
will be removed at some point.
These messages always start with [warning]
and can span multiple lines.
If there are many messages being produced --limit_output <num_messages>
can be used to limit the number of log messages produced.
[warning] NAN prediction in example 1, forcing 0
A description of audit
can be found here. These messages currently only go to stdout
.
Predictions which VW produces can be output with --predictions <output_file>
. If stdout
is passed, the results will be printed to stdout
, otherwise they are written to a file with that name.
- Home
- First Steps
- Input
- Command line arguments
- Model saving and loading
- Controlling VW's output
- Audit
- Algorithm details
- Awesome Vowpal Wabbit
- Learning algorithm
- Learning to Search subsystem
- Loss functions
- What is a learner?
- Docker image
- Model merging
- Evaluation of exploration algorithms
- Reductions
- Contextual Bandit algorithms
- Contextual Bandit Exploration with SquareCB
- Contextual Bandit Zeroth Order Optimization
- Conditional Contextual Bandit
- Slates
- CATS, CATS-pdf for Continuous Actions
- Automl
- Epsilon Decay
- Warm starting contextual bandits
- Efficient Second Order Online Learning
- Latent Dirichlet Allocation
- VW Reductions Workflows
- Interaction Grounded Learning
- CB with Large Action Spaces
- CB with Graph Feedback
- FreeGrad
- Marginal
- Active Learning
- Eigen Memory Trees (EMT)
- Element-wise interaction
- Bindings
-
Examples
- Logged Contextual Bandit example
- One Against All (oaa) multi class example
- Weighted All Pairs (wap) multi class example
- Cost Sensitive One Against All (csoaa) multi class example
- Multiclass classification
- Error Correcting Tournament (ect) multi class example
- Malicious URL example
- Daemon example
- Matrix factorization example
- Rcv1 example
- Truncated gradient descent example
- Scripts
- Implement your own joint prediction model
- Predicting probabilities
- murmur2 vs murmur3
- Weight vector
- Matching Label and Prediction Types Between Reductions
- Zhen's Presentation Slides on enhancements to vw
- EZExample Archive
- Design Documents
- Contribute: