Vowpal Wabbit is a machine learning system which pushes the frontier of machine learning with techniques

fixed Runtime library mismatch between zlib, libvw, vw.exe, VowpalWabbitCore.dll (CLR),... by using zlib/boost nuget provided msbuild targets included Visual Leak Detector for memory leak detection on windows refactored C# API to allow users to dynamically constructor serializers based on alternate descriptions (not just on static annotations) string marshalling is compatible to command line (either escaping or splitting) schema based pre-hashing: if hash can be determine from schema it's only generated once and re-used for each example. added type extension API for marshalling allow user to generate native and string examples in parallel in both debug and release keep marshalling expression tree for debugging refactored marshalling expression tree generation to improve readability added sweeping helper improved C# label parsing extensibility added assembly signing fixed memory leaks in C# usage of VW fixed model hashing/reload interaction fixed handling of empty line examples within set of action dependent features fixed order issue when predicting ADF examples containing empty action dependent features fixed default namespace incompatibility (space vs. 0) improved RunTests to C# test wrapping (detects inter-test dependencies and input files) unit tests are run in test/ folder, thus no need copy all input files added user-supplied model id support

Describe the bug

VWRegressor provides very different performance for loss_function = 'quantile' , quantile_tau = 0.5 and loss_function = 'squared'

loss_function = 'squared' - provides very GOOD low MAE loss_function = 'quantile' , quantile_tau = 0.5 - provides bad high MAE

data is mixture categorical data and continues data : 600 rows like this

To Reproduce

        if 1:
            model = VWRegressor(convert_to_vw = False ,normalized = True, 
                                                           passes = passes, 
                                                            power_t = 0.5, #1.0,
                                                           readable_model = 'my_VW.model' , cache_file =  'my_VW.cache' ,
                                                           learning_rate = 2.3 , l2 = l2, l1=l1,
                                                           quadratic= 'CC' , cubic = 'CCC',
                                                            loss_function = 'quantile' , quantile_tau = 0.5)
            q=0
        else:
            model = VWRegressor(convert_to_vw = False ,normalized = True, 
                                                      passes = passes, 
                                                       power_t = 0.5, #1.0,
                                                      readable_model = 'my_VW.model' , cache_file =  'my_VW.cache' ,
                                                      learning_rate = 2.1, loss_function = 'squared' , l2 = l2, l1=l1,
                                                      quadratic= 'CC' , cubic = 'CCC' )

Expected behavior

my guess MAE for loss_function = 'quantile' , quantile_tau = 0.5 and loss_function = 'squared' should be very similar

in addition loss_function = 'quantile' , quantile_tau = 0.9 and loss_function = 'quantile' , quantile_tau = 0.1 gives very wide confidence intervals - even no sense confidence intervals

Observed Behavior

How did VW behave? Please include any stack trace, log messages or crash logs.

Environment

What version of VW did you use? latest OS - windows 10

Additional context

do you have code example where VWRegressor loss_function = 'quantile' , quantile_tau = 0.9 and loss_function = 'quantile' , quantile_tau = 0.1

Problem

In the JNI layer, when multiple passes are enabled (> 1) and an attempt is made to close separate models in parallel, exceptions can be thrown. This is true even though each model has it's own lock to guard all accesses to the native code paths. Only a global lock around calls to the model close methods seems to avoid this issue.

I'm looking for help on identifying if any critical sections of the C code that can be guarded by a lock to avoid thread-safety issues. I'm not asking for the C code to lock. I just want help trying to figure out where to put the locks in the Java code that wraps the C code.

Scope

This seems to be in PR #1291 but was not fixed with PR #1295

Discussion

From empirical testing, it appears one of these lines seems to be the problem. I am wondering if any of these use global state. I am trying to figure out if we can lock only over a short critical section to avoid thread-safety issues.

1. adjust_used_index(*vwInstance);
2. vwInstance->do_reset_source = true;
3. VW::start_parser(*vwInstance);
4. LEARNER::generic_driver(*vwInstance);
5. VW::end_parser(*vwInstance);

Previous Conversation

In PR #1295 there was the following conversation:

@JohnLangford

There should be zero shared state between multiple created VW objects. Is that what it's doing? (Creating multiple distinct VW objects?)

@deaktator

@JohnLangford. It looks like just one VW object. Each Java call does the following on the C side:

vw* vwInstance = VW::initialize(env->GetStringUTFChars(command, NULL);

@JohnLangford

A single VW object can not be operated on in multiple threads because the code inside VW is not thread safe. If you want to have a model which is shared by multiple threads, you set this up more explicitly by initializing a new VW object with an existing model.

@deaktator

Hey @JohnLangford. We take care of multi-threaded access to VW by locking anywhere that requires access to the C code. The thread-safety issues I encountered before were on an incomplete version of the code that locked in the wrong place. When I run the tests in parallel, they seem to work just fine now. I ran them a bunch of times with forking in the tests and didn't see any issues.

Tracking Down What's Happening

It appears @jon-morra-zefr pretty much copied the C# code for multiple passes, so this seems like it might apply to C# as well. Both C# and JNI C++ code appear below as well as the calling code that blows up.

I've seen a bunch of different errors that occur at the same spot. invalidated cache, malformed LDF feature exceptions, etc.

Example Code That triggers exceptions

// Doing this many times in parallel with no locks causes problems.

	val vwJNI = VWLearners.create[VWTypedLearner[_]](vwLearnString)
	// Learning in here using   vwJNI.learn

	// PROBLEM AREA:
	lock.lock()     // <== NEED GLOBAL LOCKING OR EXCEPTIONS THROWN
	vwJNI.close()
	lock.unlock()   // <== NEED GLOBAL LOCKING OR EXCEPTIONS THROWN

Similarity of the C# and JNI C++ Code

C# Code: vowpal_wabbit/cs/cli/vowpalwabbit.cpp

void VowpalWabbit::RunMultiPass()
{ if (m_vw->numpasses > 1)
  { try
    { adjust_used_index(*m_vw);
      m_vw->do_reset_source = true;
      VW::start_parser(*m_vw);
      LEARNER::generic_driver(*m_vw);
      VW::end_parser(*m_vw);
    }
    CATCHRETHROW
  }
}

JNI C++ Code: vowpal_wabbit/java/src/main/c++/vowpalWabbit_learner_VWLearners.cc

JNIEXPORT void JNICALL Java_vowpalWabbit_learner_VWLearners_performRemainingPasses(JNIEnv *env, jclass obj, jlong vwPtr)
{ try
  { vw* vwInstance = (vw*)vwPtr;
    if (vwInstance->numpasses > 1)
      { adjust_used_index(*vwInstance);
        vwInstance->do_reset_source = true;
        VW::start_parser(*vwInstance);
        LEARNER::generic_driver(*vwInstance);
        VW::end_parser(*vwInstance);
      }
  }
  catch(...)
  { rethrow_cpp_exception_as_java_exception(env);
  }
}

Any thoughts?

This is the preliminary PR for continuous actions.

This includes, cats_tree (continuous action tree with smoothing) algorithm, converting between PMF (discrete) to PDF (continuous) distribution, sampling form continuous PDF, etc.

The code is for the paper available at https://arxiv.org/pdf/2006.06040.pdf

We will add more details.

Fixed NRE on empty hashes Skip model load/initialize when seeding from in-memory model Fixed progressive validation in Azure trainer Includes mixed JSON string and JSON direct support Includes native C++ JSON parsing

Added 3 new make targets: vw_gcov, library_example_gcov, test_gcov which build vw and the examples with GCOV support, then run tests. This allows coveralls to analyze test coverage in the source code, but slows the tests down signifigantly. I also edited the travis .yml file to upload the results to coveralls.io and added the badge to the readme.

Someone will need to setup a coveralls account for the main VW project and point the badge in the readme to that badge. Currently the coveralls badge points only to my fork.

1. Overview

The goal of this PR is to fix the issue #2308.

The PR introduces a new class DFToVW in vowpalwabbit.pyvw that takes as input the pandas.DataFrame and special types (SimpleLabel, Feature, Namespace) that specify the desired VW conversion.

These classes make extensive use of a class Col that refers to a given column in the user specified dataframe.

A simpler interface DFtoVW.from_colnames also be used for the simple use-cases. The main benefit is that the user need not use the specific types.

Below are some usages of this class. They all rely on the following pandas.DataFrame called df :

  house_id  need_new_roof  price  sqft   age  year_built
0      id1              0   0.23  0.25  0.05        2006
1      id2              1   0.18  0.15  0.35        1976
2      id3              0   0.53  0.32  0.87        1924

2. Simple usage using DFtoVW.from_colnames

Let say we want to build a VW dataset with the target need_new_roof and the feature age :

from vowpalwabbit.pyvw import DFtoVW
conv = DFtoVW.from_colnames(y="need_new_roof", x=["age", "year_built"], df=df)

Then we can use the method process_df:

conv.process_df()

that outputs the following list:

['0 | 0.05 2006', '1 | 0.35 1976', '0 | 0.87 1924']

This list can then directly be consumed by the method pyvw.model.learn.

3. Advanced usages using default constructor

The class DFtoVW also allow the following patterns in its default constructor :

• tag
• (named) namespaces, with scaling factor
• (named) features, with constant feature possible

To use these more complex patterns we need to import them using:

from vowpalwabbit.pyvw import SimpleLabel, Namespace, Feature, Col

3.1. Named namespace with scaling, and named feature

Let's create a VW dataset that include a named namespace (with scaling) and a named feature:

conv = DFtoVW(
        df=df,
        label=SimpleLabel(Col("need_new_roof")),
        namespaces=Namespace(name="Imperial", value=0.092, features=Feature(value=Col("sqft"), name="sqm"))
        )
conv.process_df()

which yields:

['0 |Imperial:0.092 sqm:0.25',
 '1 |Imperial:0.092 sqm:0.15',
 '0 |Imperial:0.092 sqm:0.32']

3.2. Multiple namespaces, multiple features, and tag

Let's create a more complex example with a tag and multiples namespaces with multiples features.

conv = DFtoVW(
        df=df, 
        label=SimpleLabel(Col("need_new_roof")),
        tag=Col("house_id"),
        namespaces=[
                Namespace(name="Imperial", value=0.092, features=Feature(value=Col("sqft"), name="sqm")),
                Namespace(name="DoubleIt", value=2, features=[Feature(value=Col("price")), Feature(Col("age"))])
                ]
        )
conv.process_df()

which yields:

['0 id1|Imperial:0.092 sqm:0.25 |DoubleIt:2 0.23 0.05',
 '1 id2|Imperial:0.092 sqm:0.15 |DoubleIt:2 0.18 0.35',
 '0 id3|Imperial:0.092 sqm:0.32 |DoubleIt:2 0.53 0.87']

4. Implementation details

• The class DFtoVW and the specific types are located in vowpalwabbit/pyvw.py. The class only depends on the pandas module.
• the code includes docstrings
• 8 tests are included in tests/test_pyvw.py

5. Extensions

• This PR does not yet handle multilines and more complex label types.
• To convert very large dataset that can't fit in RAM, one can make use of the pandas import option chunksize and process each chunk at a time. I could also implement this functionnality directly in the class using generator. The generator would then be consumed by either a VW learning interface or could be written to external file (for conversion purpose only).

Mac os 10.10 and boost version boost-1.58.0

gcc --version Configured with: --prefix=/Library/Developer/CommandLineTools/usr --with-gxx-include-dir=/usr/include/c++/4.2.1 Apple LLVM version 6.1.0 (clang-602.0.53) (based on LLVM 3.6.0svn) Target: x86_64-apple-darwin14.0.0 Thread model: posix

Get some warnings and test 16 don't pass.

Here is full log:

make
cd vowpalwabbit; /Library/Developer/CommandLineTools/usr/bin/make -j 4 things
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c main.cc -o main.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c global_data.cc -o global_data.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c parse_regressor.cc -o parse_regressor.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c parse_primitives.cc -o parse_primitives.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c unique_sort.cc -o unique_sort.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cache.cc -o cache.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c rand48.cc -o rand48.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c simple_label.cc -o simple_label.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c multiclass.cc -o multiclass.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c oaa.cc -o oaa.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c multilabel_oaa.cc -o multilabel_oaa.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c boosting.cc -o boosting.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c ect.cc -o ect.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c autolink.cc -o autolink.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c binary.cc -o binary.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c lrq.cc -o lrq.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cost_sensitive.cc -o cost_sensitive.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c multilabel.cc -o multilabel.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c label_dictionary.cc -o label_dictionary.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c csoaa.cc -o csoaa.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cb.cc -o cb.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cb_adf.cc -o cb_adf.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cb_algs.cc -o cb_algs.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search.cc -o search.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_meta.cc -o search_meta.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_sequencetask.cc -o search_sequencetask.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_dep_parser.cc -o search_dep_parser.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_hooktask.cc -o search_hooktask.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_multiclasstask.cc -o search_multiclasstask.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_entityrelationtask.cc -o search_entityrelationtask.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c search_graph.cc -o search_graph.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c parse_example.cc -o parse_example.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c scorer.cc -o scorer.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c network.cc -o network.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c parse_args.cc -o parse_args.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c accumulate.cc -o accumulate.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c gd.cc -o gd.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c learner.cc -o learner.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c lda_core.cc -o lda_core.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c gd_mf.cc -o gd_mf.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c mf.cc -o mf.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c bfgs.cc -o bfgs.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c noop.cc -o noop.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c print.cc -o print.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c example.cc -o example.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c parser.cc -o parser.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c loss_functions.cc -o loss_functions.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c sender.cc -o sender.o
parser.cc:452:26: warning: 'daemon' is deprecated: first deprecated in OS X 10.5
      [-Wdeprecated-declarations]
      if (!all.active && daemon(1,1))
                         ^
/usr/include/stdlib.h:267:6: note: 'daemon' has been explicitly marked
      deprecated here
int      daemon(int, int) __DARWIN_1050(daemon) __OSX_AVAILABLE_BUT_DEPR...
         ^
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c nn.cc -o nn.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c bs.cc -o bs.o
1 warning generated.
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c cbify.cc -o cbify.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c topk.cc -o topk.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c stagewise_poly.cc -o stagewise_poly.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c log_multi.cc -o log_multi.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c active.cc -o active.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c kernel_svm.cc -o kernel_svm.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c best_constant.cc -o best_constant.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c ftrl.cc -o ftrl.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c svrg.cc -o svrg.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c lrqfa.cc -o lrqfa.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c interact.cc -o interact.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c comp_io.cc -o comp_io.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c interactions.cc -o interactions.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c vw_exception.cc -o vw_exception.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c allreduce.cc -o allreduce.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o active_interactor active_interactor.cc
ar rcs liballreduce.a allreduce.o
ar rcs libvw.a hash.o global_data.o io_buf.o parse_regressor.o parse_primitives.o unique_sort.o cache.o rand48.o simple_label.o multiclass.o oaa.o multilabel_oaa.o boosting.o ect.o autolink.o binary.o lrq.o cost_sensitive.o multilabel.o label_dictionary.o csoaa.o cb.o cb_adf.o cb_algs.o search.o search_meta.o search_sequencetask.o search_dep_parser.o search_hooktask.o search_multiclasstask.o search_entityrelationtask.o search_graph.o parse_example.o scorer.o network.o parse_args.o accumulate.o gd.o learner.o lda_core.o gd_mf.o mf.o bfgs.o noop.o print.o example.o parser.o loss_functions.o sender.o nn.o bs.o cbify.o topk.o stagewise_poly.o log_multi.o active.o kernel_svm.o best_constant.o ftrl.o svrg.o lrqfa.o interact.o comp_io.o interactions.o vw_exception.o
/usr/bin/g++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o vw main.o -L. -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
cd cluster; /Library/Developer/CommandLineTools/usr/bin/make
/usr/bin/clang++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -c spanning_tree.cc -o spanning_tree.o
spanning_tree.cc:161:9: warning: 'daemon' is deprecated: first deprecated in OS
      X 10.5 [-Wdeprecated-declarations]
    if (daemon(1,1))
        ^
/usr/include/stdlib.h:267:6: note: 'daemon' has been explicitly marked
      deprecated here
int      daemon(int, int) __DARWIN_1050(daemon) __OSX_AVAILABLE_BUT_DEPR...
         ^
1 warning generated.
/usr/bin/clang++ -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o spanning_tree spanning_tree.o 
cd library; /Library/Developer/CommandLineTools/usr/bin/make things
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o ezexample_predict ezexample_predict.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o ezexample_train ezexample_train.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o library_example library_example.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o recommend recommend.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o gd_mf_weights gd_mf_weights.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z
/usr/bin/g++ -g -std=c++0x   -D__extern_always_inline=inline -Wall -pedantic -O3 -fomit-frame-pointer -fno-strict-aliasing  -D_FILE_OFFSET_BITS=64 -DNDEBUG -I /usr/local/include  -fPIC  -o test_search test_search.cc -L ../vowpalwabbit -l vw -l allreduce -L /usr/local/lib -lboost_program_options-mt -lboost_serialization-mt -l pthread -l z

make test
cd vowpalwabbit; /Library/Developer/CommandLineTools/usr/bin/make -j 4 things
make[1]: Nothing to be done for `things'.
cd library; /Library/Developer/CommandLineTools/usr/bin/make things
make[1]: Nothing to be done for `things'.
vw running test-suite...
(cd test && ./RunTests -d -fe -E 0.001 ../vowpalwabbit/vw ../vowpalwabbit/vw)
Testing on: hostname=air-mac OS=darwin
Testing vw: ../vowpalwabbit/vw
Testing lda: ../vowpalwabbit/vw
RunTests: '-D' to see any diff output
RunTests: '-o' to force overwrite references
RunTests: test 1: stderr OK
RunTests: test 2: stderr OK
RunTests: test 2: predict OK
RunTests: test 3: stderr OK
RunTests: test 4: stdout OK
RunTests: test 4: stderr OK
RunTests: test 5: stderr OK
RunTests: test 6: stderr OK
RunTests: test 6: minor (<0.001) precision differences ignored
RunTests: test 6: predict OK
RunTests: test 7: stderr OK
RunTests: test 8: stderr OK
RunTests: test 8: minor (<0.001) precision differences ignored
RunTests: test 8: predict OK
RunTests: test 9: stderr OK
RunTests: test 9: predict OK
RunTests: test 10: stderr OK
RunTests: test 10: predict OK
RunTests: test 11: stderr OK
RunTests: test 12: stderr OK
RunTests: test 13: stderr OK
RunTests: test 14: stdout OK
RunTests: test 14: minor (<0.001) precision differences ignored
RunTests: test 14: stderr OK
RunTests: test 15: stdout OK
RunTests: test 15: stderr OK
RunTests: test 16: stdout OK
--- diff -u --minimal train-sets/ref/rcv1_small.stderr stderr.tmp
--- train-sets/ref/rcv1_small.stderr    2015-08-13 00:22:20.000000000 +0300
+++ stderr.tmp  2015-08-13 00:33:33.000000000 +0300
@@ -17,7 +17,7 @@
  5 0.47879     0.00006     0.00617      0.595892   0.183063                            0.47184     1.00000   
  6 0.47750     0.00000     0.00221      0.703360   0.403715                            0.68626     1.00000   
  7 0.47680     0.00000     0.00038      0.588395   0.175459                            0.08911     1.00000   
- 8 0.47671     0.00000     0.00002      0.568445   0.136827                            0.00444     1.00000   
+ 8 0.47671     0.00000     0.00002      0.568443   0.136827                            0.00444     1.00000   

 finished run
 number of examples = 8000
RunTests: test 16: FAILED: ref(train-sets/ref/rcv1_small.stderr) != stderr(stderr.tmp)
    cmd: ../vowpalwabbit/vw -k -c -d train-sets/rcv1_small.dat --loss_function=logistic -b 20 --bfgs --mem 7 --passes 20 --termination 0.001 --l2 1.0 --holdout_off

Using VW 8.4.0 installed with brew I created a simple test set initializing VW with

$ vw --csoaa 10  -b 24  --l2 0.0  -l 0.1  -c -k --passes 100  -f /Users/pat/big-data/harness/models/test_resource  --save_resume

Then I paste examples in:

0:0.0 1:1.0 | user_user_2 testGroupId_1 
0:0.0 1:1.0 | user_user_2 testGroupId_1 
0:0.0 1:1.0 | user_user_2 testGroupId_1 
0:0.0 1:1.0 | user_user_2 testGroupId_1 
0:1.0 1:0.0 | user_user_1 testGroupId_1 
0:1.0 1:0.0 | user_user_1 testGroupId_1 
0:1.0 1:0.0 | user_user_1 testGroupId_1 
0:1.0 1:0.0 | user_user_1 testGroupId_1 
save_

at the save_ the file Users/pat/big-data/harness/models/test_resource is updated—all is well.

Using the last available JNI binary wrapper for 8.2.0 doing the same thing from Java does not update the model file. I'm not running in --quiet mode and there is no complaint from VW.

The save_ pseudo example does apparently work for 8.2.0 since another user is using it in CLI and daemon mode.

Is this feature not supported with JNI?

On the advice of @arielf it appears I need something like vw-jni-8.4.1-SNAPSHOT so trying to build for dev machine (MBP) and deploy machine (ubuntu). Dev machine first.

I'm not able to pip install vowpalwabbit to install the python wrapper. I don't know enough to understand why it's failing, but I thought it might be worth bringing to someone's attention.

I'm on OSX and using an Anaconda environment. I installed vowpal wabbit from homebrew.

Here's my traceback:

Collecting vowpalwabbit
  Using cached vowpalwabbit-8.2.0.tar.gz
Building wheels for collected packages: vowpalwabbit
  Running setup.py bdist_wheel for vowpalwabbit ... error
  Complete output from command /Users/vvvvv/anaconda/envs/trendrank/bin/python -u -c "import setuptools, tokenize;__file__='/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py';exec(compile(getattr(tokenize, 'open', open)(__file__).read().replace('\r\n', '\n'), __file__, 'exec'))" bdist_wheel -d /var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/tmp5FosrOpip-wheel- --python-tag cp27:
  running bdist_wheel
  running build
  running build_py
  creating build
  creating build/lib.macosx-10.5-x86_64-2.7
  creating build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
  copying vowpalwabbit/__init__.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
  copying vowpalwabbit/pyvw.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
  copying vowpalwabbit/sklearn_vw.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
  running egg_info
  writing vowpalwabbit.egg-info/PKG-INFO
  writing top-level names to vowpalwabbit.egg-info/top_level.txt
  writing dependency_links to vowpalwabbit.egg-info/dependency_links.txt
  warning: manifest_maker: standard file '-c' not found

  reading manifest file 'vowpalwabbit.egg-info/SOURCES.txt'
  reading manifest template 'MANIFEST.in'
  warning: no previously-included files matching '*.o' found anywhere in distribution
  warning: no previously-included files matching '*.exe' found anywhere in distribution
  warning: no previously-included files matching '*.pyc' found anywhere in distribution
  writing manifest file 'vowpalwabbit.egg-info/SOURCES.txt'
  running build_ext
  Traceback (most recent call last):
    File "<string>", line 1, in <module>
    File "/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py", line 184, in <module>
      tests_require=['tox'],

[...]

    File "/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py", line 38, in find_boost
      raise Exception('Could not find boost python library')
  Exception: Could not find boost python library

  ----------------------------------------
  Failed building wheel for vowpalwabbit
  Running setup.py clean for vowpalwabbit
Failed to build vowpalwabbit
Installing collected packages: vowpalwabbit
  Running setup.py install for vowpalwabbit ... error
    Complete output from command /Users/vvvvv/anaconda/envs/trendrank/bin/python -u -c "import setuptools, tokenize;__file__='/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py';exec(compile(getattr(tokenize, 'open', open)(__file__).read().replace('\r\n', '\n'), __file__, 'exec'))" install --record /var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-4GRRiq-record/install-record.txt --single-version-externally-managed --compile:
    running install
    running build
    running build_py
    creating build
    creating build/lib.macosx-10.5-x86_64-2.7
    creating build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
    copying vowpalwabbit/__init__.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
    copying vowpalwabbit/pyvw.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
    copying vowpalwabbit/sklearn_vw.py -> build/lib.macosx-10.5-x86_64-2.7/vowpalwabbit
    running egg_info
    creating vowpalwabbit.egg-info
    writing vowpalwabbit.egg-info/PKG-INFO
    writing top-level names to vowpalwabbit.egg-info/top_level.txt
    writing dependency_links to vowpalwabbit.egg-info/dependency_links.txt
    writing manifest file 'vowpalwabbit.egg-info/SOURCES.txt'
    warning: manifest_maker: standard file '-c' not found

    reading manifest file 'vowpalwabbit.egg-info/SOURCES.txt'
    reading manifest template 'MANIFEST.in'
    warning: no files found matching '*' under directory 'src'
    warning: no previously-included files matching '*.o' found anywhere in distribution
    warning: no previously-included files matching '*.exe' found anywhere in distribution
    warning: no previously-included files matching '*.pyc' found anywhere in distribution
    writing manifest file 'vowpalwabbit.egg-info/SOURCES.txt'
    running build_ext
    make: *** No rule to make target `clean'.  Stop.
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py", line 184, in <module>
        tests_require=['tox'],

[...]

    subprocess.CalledProcessError: Command '['make', 'clean']' returned non-zero exit status 2

    ----------------------------------------
Command "/Users/vvvvv/anaconda/envs/trendrank/bin/python -u -c "import setuptools, tokenize;__file__='/private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/setup.py';exec(compile(getattr(tokenize, 'open', open)(__file__).read().replace('\r\n', '\n'), __file__, 'exec'))" install --record /var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-4GRRiq-record/install-record.txt --single-version-externally-managed --compile" failed with error code 1 in /private/var/folders/vx/n21m77w12nd0tb9xwhfcgd75gpm86h/T/pip-build-74A7hN/vowpalwabbit/

This is a pretty major refactoring of the JNI layer. The impetus for this refactoring was the ability to accept multiline examples but it lead to a much larger change. The biggest change is the decoupling of the return type and the prediction function. This proved necessary to support all the different ways to extract data with --cb_explore.

I am going to go over this in detail with @deaktator offline, so @JohnLangford let's hold off on merging this for now. If anyone else has any comments at this time they are certainly welcome.

I am essentially just stabilizing the initialize_experimental function which has been working well for quite some time. With the changes of adding skip_model_load and removing the custom stack functionality from the stable interface. The custom stack is still available from initialize_experimental

std::unique_ptr<VW::workspace> initialize(std::unique_ptr<config::options_i> options,
    std::unique_ptr<VW::io::reader> model_override_reader = nullptr, bool skip_model_load = false,
    driver_output_func_t driver_output_func = nullptr, void* driver_output_func_context = nullptr,
    VW::io::logger* custom_logger = nullptr);

Forcing all consumers to have a preallocated pool is quite restrictive, especially now that the pool is dynamic (in the past it was a fixed size ring buffer).

Removing this initial allocation improves support for library scenarios, and in practice should have no effect on driver performance. (hyperfine shows this too)

This change from a default of 256 to an initial size of 0 means that base memory consumption is halved (An example is ~31kb) from 13.5MB to 6.2MB.

Startup should also be reduced. While it is hard to measure: vw --no_stdin using hyperfine shows:

Benchmark 1: ./build/vowpalwabbit/cli/vw --no_stdin
  Time (mean ± σ):       4.5 ms ±   2.1 ms    [User: 1.6 ms, System: 0.9 ms]
  Range (min … max):     3.0 ms …  30.6 ms    682 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet PC without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Benchmark 2: ./vw-master --no_stdin
  Time (mean ± σ):       5.8 ms ±   1.6 ms    [User: 2.0 ms, System: 1.5 ms]
  Range (min … max):     4.2 ms …  21.6 ms    658 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet PC without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Summary
  './build/vowpalwabbit/cli/vw --no_stdin' ran
    1.29 ± 0.69 times faster than './vw-master --no_stdin'

Describe the bug

Hello,

I am currently working on the conditinal contextual combinatorial bandit, i.e. ccb. I have 4 slots and a total of 12 actions. I am training my algorithm with the following code:

def formatting_and_training_bandit_input_data(bandit, vowpal_workspace):
    for i in bandit.Id.unique():
        bandit_loop =bandit[bandit['Id'] == i]
        bandit_loop.reset_index(inplace = True,drop = True)
        action =bandit_loop.Action
        prob_action =bandit_loop.prob_action
        reward=bandit_loop.Reward * -100
        userid = bandit_loop.UserId
        gender = bandit_loop.Gender
        age =bandit_loop.Age
        daten =[
        f"ccb shared  | UserId={userid[0]} Gender={gender[0]}  Age={age[0]} clicks_total:{bandit_loop.clicks_total[0]} AmountPerOrder_Median_D365:{bandit_loop.AmountPerOrder_Median_D365[0]} avg_product_amount:{bandit_loop.avg_product_amount[0]}  day_lenght:{bandit_loop.day_lenght[0]}",        "ccb action |",
        "ccb action | 0",
        "ccb action | 1",
        "ccb action | 2",
        "ccb action | 3",
        "ccb action | 4",
        "ccb action | 5",
        "ccb action | 6",
        "ccb action | 7",
        "ccb action | 8",
        "ccb action | 9",
        "ccb action | 10",
        "ccb action | 11",
        f"ccb slot {action.iloc[0 ]}:{prob_action.iloc[0 ]}:{reward.iloc[0]} |",
        f"ccb slot {action.iloc[1]}:{prob_action.iloc[1 ]}:{reward.iloc[1] } |",
        f"ccb slot {action.iloc[2]}:{prob_action.iloc[2 ]}:{reward.iloc[ 2]} |",
        f"ccb slot {action.iloc[3]}:{prob_action.iloc[3 ]}:{reward.iloc[3]} |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        # f"ccb slot |",
        ]
            
       
        vowpal_workspace.learn(daten)
    vowpal_workspace.save("cb.saved")
    return daten

I would like to make the forecast with the following data:

def formatting_and_predicting_bandit_input_data(bandit, vowpal_workspace):
    prediction_result = []
    for i in bandit.Id.unique():
        bandit_loop =bandit[bandit['Id'] == i]
        bandit_loop.reset_index(inplace = True,drop = True)
        # action =bandit_loop.Action
        # reward=bandit_loop.Reward * -1
        userid = bandit_loop.UserId
        gender = bandit_loop.Gender
        age =bandit_loop.Age
        daten =[
        f"ccb shared  | UserId={userid[0]} Gender={gender[0]}  Age={age[0]} clicks_total:{bandit_loop.clicks_total[0]} AmountPerOrder_Median_D365:{bandit_loop.AmountPerOrder_Median_D365[0]} avg_product_amount:{bandit_loop.avg_product_amount[0]}  day_lenght:{bandit_loop.day_lenght[0]}",
        "ccb action | 0",
        "ccb action | 1",
        "ccb action | 2",
        "ccb action | 3",
        "ccb action | 4",
        "ccb action | 5",
        "ccb action | 6",
        "ccb action | 7",
        "ccb action | 8",
        "ccb action | 9",
        "ccb action | 10",
        "ccb action | 11",
        f"ccb slot |",
        f"ccb slot |",
        f"ccb slot |",
        f"ccb slot |",

        ]
            

        pred= vowpal_workspace.predict(daten)
        prediction_result.append(pred[0][0][0])
        prediction_result.append(pred[1][0][0])
        prediction_result.append(pred[2][0][0])
        prediction_result.append(pred[3][0][0])
    return prediction_result

The problem is that the forecast is not good. All 12 actions are used with the same frequency. Can anyone help me here? I

How to reproduce

you simply cant

Version

22

OS

Ubuntu

Language

Python

Additional context

Conditional Contextual Bandit