Hivemall v0.3.2-3 (maintenance release)

This is just a maintenance release of Hivemall v0.3.2.
Only minor changes have been applied. A support for efficient Top-K query processing is the main update in this release.

---

• Major Enhancement
  • Added each_top_k UDTF [10ed705]
• Minor Enhancement
  • Added x_rank() function [333d3a6]
  • Added euclid_similarity UDF [10d6e23]
  • Added manhattan_distance/minkowski_distance UDFs and distance2similarity UDF [821060a]
• Minor Changes
  • Added no_bias option to Matrix Factorization [c94f993]
  • Modified distance/similarity UDF to subclasses of GenericUDF for a better performance [29f1102]

Hivemall v0.3.2-2 (maintenance release)

This is just a maintenance release of Hivemall v0.3.2.
Only minor changes have been applied.

---

• Minor Changes
  • Supported -no_bias option in Matrix Factorization [d6d2ac2]
  • Fixed to accept BigInt type for label in Multi-class classification [0c79c82]
  • Changed categorical_features() behavior [00cc200]

Hivemall v0.3.2-1 (maintenance release)

This is just a maintenance release of Hivemall v0.3.2.

Only minor changes have been applied since the last release as follows:

• Minor Enhancement
  • Added feature(name, value) UDF [ea5e089]
  • Added jaccard_distance UDF [adcf2f8]
  • Added categorical_feature UDF [86d3011]
• Minor Changes
  • Updated jaccard to accept arrays as the inputs [ea5e089]
  • Fixed to_string_array to accept ANY primitive types [1860636]

The release version of Hivemall v0.3.2

This is the stable release version of Hivemall v0.3.2.

A major enhancement in this release is support for anomaly detection using LOF and polynomial features that is useful for non-linear regression/classification.

Changes since v0.3.1 are summarized as follows:

• Major Enhancement
  • Supported Local Outlier Factor (LOF) [0d32487]
  • Added polynomial_feature and powered_feature UDFs [95c60f0][cf1aacc]
• Minor Enhancement
  • Modified to accept double values to avoid annoying casts in rescale/zscore [1485d7e]
  • Added to_string_array UDF that convert array() to array() [8b910a6]
  • Added euclid_distance function [21d4d9a]
  • Modified to accept various numeric types for userid and itemid in Matrix Factorization [078cd85]
  • Added cosine_distance UDF [1f8f127]
  • Added angular_similarity and angular_distance UDFs [00ff531]
  • Added vectorize_features UDF [3cedd66][f165d49]
• Minor Changes
  • Added alias cosine_similarity to cosine_sim [3f615e5]
  • Modified to return null instead of throwing UDFArgumentException for a null argument in mhash [181b369]
  • Moved the package of similarity UDFs from hivemall.knn.distance to hivemall.knn.similarity and changed the function signature of consine_similarity function. [2a0f1e7]
  • Changed argument signatures from INT to DOUBLE in TF-IDF macro [61eeca2]
• Bugfix
  • Fixed a bug in java_min [a48c367]

The release version of Hivemall v0.3.1

This is the stable release version of Hivemall v0.3.1. We have changed the license of Hivemall from LGPL v2 to Apache License v2 from this release.

From this release, stable releases of Hivemall are also released on Maven Central.

Changes since v0.3 are summarized as follows:

• Major Enhancement
  • Supported Hivemall on Pig (on Pig v0.15 or later) [8aac879]
    Thanks Daniel for the contribution!
  • Released Hivemall on Maven Central

<dependency>
    <groupId>io.github.myui</groupId>
    <artifactId>hivemall</artifactId>
    <version>0.3.1</version>
</dependency>

• Minor Enhancement
  • Added add_feature_index() UDF [bad68ec]
  • Added hivemall_version() UDF [ae493a2]

-- assign indices to dense features
select add_feature_index(array(3,4.0,5)) from dual;
> ["1:3.0","2:4.0","3:5.0"]

• Major Changes
  • Changed the license from LGPL v2 to Apache License v2 [cc8be7e]
• Minor Changes
  • Modifed extract_weight to accept a categorical feature representation [f5a3c28]
  • Modified to accept various types in addition to INT for label (classification) [c352670]
  • Modified to accept both float/double in label/target (regression) [35bedf1]
  • Feature parsing scheme has been refactored to be more efficient [e37c47f]
  • Changed the implementation of array_avg from ArrayAvgUDAF to ArrayAvgGenericUDAF [3a21853]

-- accept categorical variables (e.g., "weight") in Hivemall v0.3.1 in addition to quantitive variables (e.g., "weight:55.0") in extract_weight()
select extract_weight("weight"), extract_weight("weight:55.0") from dual;
> 1.0 | 55.0

select 
   -- logress(addBias(features), CAST(label as FLOAT)) as (feature, weight) -- Hivemall v0.3 (need to cast labels)
   logress(addBias(features), label) as (feature, weight) -- Hivemall v0.3.1 or later (no need to cast labels)

• Bugfixes
  • Fixed a bug in jaccard() assuming the result of dividing integer expressions as float [4115913]

The release version of Hivemall v0.3.0

This is the first stable release version of Hivemall v0.3.0.
A major enhancement within this release is support for matrix factorization.
Note that hivemall v0.3 or later supports Hive v0.11 or later.

Changes since v0.3_beta3 are summarized as follows:

• Major Enhancement
  • Supported Matrix Factorization [433ddc8]
• Minor Enhancement
  • Added f1score() UDAF that represents f-measure [6dedbbf]
  • Added kld() UDF [ 41ceeb0]
  • Added FixedEtaEstimator [5dfc7ef]
  • Supported mae/mse/rmse UDAFs [8ff0cdf]
  • Added rand_gid/rand_gid2 macro [caf5266]
  • Supported a new function, train_mf_adagrad [fb4e8f2]
• Major Changes
  • Refactored parameter averaging schemes [7ff8985][71566fd]
  • Fixed the averaging scheme of covariance [ee43d2c]
• Bugfix
  • Fixed rand_amplify() to shuffle in sweepAll() where buffer is not filled [5b798f1]
  • Fixed DoubleWritable import in kld() [2c0e7d5]

Beta version #3 of v0.3

A major change in this maintenance release is the support for TF-IDF computation.

• Major changes
  • Added a support for TF-IDF computation [876b335]
  • Changed the default eta0 (initial learning rate) to 0.2 [aa92317]
  • Added a DDL to create hivemall UDFs as permanent functions [5cc77c9]
• Minor changes
  • Added a server throughput monitoring functionality using JMX [744bc8d]
  • Fixed MixServerTest to pass a TravisCI's build

Beta version #2 of v0.3

A major change in this maintenance release is the support for AdaGrad/AdaGradRDA/AdaDelta.

For the usage, see the following examples:

• https://github.com/myui/hivemall/wiki/news20-binary-classification-AdaGradRDA-AdaGrad-AdaDelta
• https://github.com/myui/hivemall/wiki/KDDCup-2012-track-2-CTR-prediction-AdaGrad-AdaDelta

• Enhancement
  • Supported AdaGrad/AdaDelta
  • Supported AdaGradRDA classification
• Major changes
  • logress_iter() is no more supported [ad10c33]
• Minor changes
  • Fixed not to use String#split because of its performance [9e0a95c]

The initial beta release of v0.3 (beta1)

We are pleased to announce the initial beta release (beta1) of Hivemall v0.3.

The major enhancement in this release is the support for model mixing.
See this page to know how to use the new feature.

You can find a brief explanation of the internal design of MIX protocol in this slide.

Note this is a pre-release and for alpha users. Any feedback is welcome!

The stable release of Hivemall v0.2

This is the stable release of Hivemall v0.2, recommended for production uses.

From this release, Hive v0.11 or later is required.

---

Changes applied since v0.2-alpha4 is small is as follows:

• Minor changes
  • Changed the package structure w.r.t. hivemall.io.* [95f4e0b]
• Bugfix
  • Fixed a bug in the "disable_halffloat" option [1c975e2]
  • Fixed maven dependencies [a8d09ec]