http://git-wip-us.apache.org/repos/asf/incubator-hivemall-site/blob/6c6a1b42/userguide/binaryclass/criteo_ffm.html ---------------------------------------------------------------------- diff --git a/userguide/binaryclass/criteo_ffm.html b/userguide/binaryclass/criteo_ffm.html new file mode 100644 index 0000000..acaec48 --- /dev/null +++ b/userguide/binaryclass/criteo_ffm.html @@ -0,0 +1,2708 @@ + +<!DOCTYPE HTML> +<html lang="" > + <head> + <meta charset="UTF-8"> + <meta content="text/html; charset=utf-8" http-equiv="Content-Type"> + <title>Field-Aware Factorization Machines · Hivemall User Manual</title> + <meta http-equiv="X-UA-Compatible" content="IE=edge" /> + <meta name="description" content=""> + <meta name="generator" content="GitBook 3.2.3"> + + + + + <link rel="stylesheet" href="../gitbook/style.css"> + + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-splitter/splitter.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-etoc/plugin.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-callouts/plugin.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-toggle-chapters/toggle.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-codeblock-filename/block.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-expandable-chapters/expandable-chapters.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-multipart/multipart.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-katex/katex.min.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-emphasize/plugin.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-highlight/website.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-search/search.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-fontsettings/website.css"> + + + + <link rel="stylesheet" href="../gitbook/gitbook-plugin-theme-api/theme-api.css"> + + + + + + + + + + + + + + + + + + + + + + + + <meta name="HandheldFriendly" content="true"/> + <meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no"> + <meta name="apple-mobile-web-app-capable" content="yes"> + <meta name="apple-mobile-web-app-status-bar-style" content="black"> + <link rel="apple-touch-icon-precomposed" sizes="152x152" href="../gitbook/images/apple-touch-icon-precomposed-152.png"> + <link rel="shortcut icon" href="../gitbook/images/favicon.ico" type="image/x-icon"> + + + <link rel="next" href="../multiclass/news20.html" /> + + + <link rel="prev" href="criteo_dataset.html" /> + + + </head> + <body> + +<div class="book"> + <div class="book-summary"> + + +<div id="book-search-input" role="search"> + <input type="text" placeholder="Type to search" /> +</div> + + + <nav role="navigation"> + + + +<ul class="summary"> + + + + + <li> + <a href="http://hivemall.incubator.apache.org/"; target="_blank" class="custom-link"><i class="fa fa-home"></i> Home</a> + </li> + + + + + <li class="divider"></li> + + + + + <li class="header">TABLE OF CONTENTS</li> + + + + <li class="chapter " data-level="1.1" data-path="../"> + + <a href="../"> + + + <b>1.1.</b> + + Introduction + + </a> + + + + </li> + + <li class="chapter " data-level="1.2" data-path="../getting_started/"> + + <a href="../getting_started/"> + + + <b>1.2.</b> + + Getting Started + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="1.2.1" data-path="../getting_started/installation.html"> + + <a href="../getting_started/installation.html"> + + + <b>1.2.1.</b> + + Installation + + </a> + + + + </li> + + <li class="chapter " data-level="1.2.2" data-path="../getting_started/permanent-functions.html"> + + <a href="../getting_started/permanent-functions.html"> + + + <b>1.2.2.</b> + + Install as permanent functions + + </a> + + + + </li> + + <li class="chapter " data-level="1.2.3" data-path="../getting_started/input-format.html"> + + <a href="../getting_started/input-format.html"> + + + <b>1.2.3.</b> + + Input Format + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="1.3" data-path="../misc/funcs.html"> + + <a href="../misc/funcs.html"> + + + <b>1.3.</b> + + List of Functions + + </a> + + + + </li> + + <li class="chapter " data-level="1.4" data-path="../tips/"> + + <a href="../tips/"> + + + <b>1.4.</b> + + Tips for Effective Hivemall + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="1.4.1" data-path="../tips/addbias.html"> + + <a href="../tips/addbias.html"> + + + <b>1.4.1.</b> + + Explicit add_bias() for better prediction + + </a> + + + + </li> + + <li class="chapter " data-level="1.4.2" data-path="../tips/rand_amplify.html"> + + <a href="../tips/rand_amplify.html"> + + + <b>1.4.2.</b> + + Use rand_amplify() to better prediction results + + </a> + + + + </li> + + <li class="chapter " data-level="1.4.3" data-path="../tips/rt_prediction.html"> + + <a href="../tips/rt_prediction.html"> + + + <b>1.4.3.</b> + + Real-time prediction on RDBMS + + </a> + + + + </li> + + <li class="chapter " data-level="1.4.4" data-path="../tips/ensemble_learning.html"> + + <a href="../tips/ensemble_learning.html"> + + + <b>1.4.4.</b> + + Ensemble learning for stable prediction + + </a> + + + + </li> + + <li class="chapter " data-level="1.4.5" data-path="../tips/mixserver.html"> + + <a href="../tips/mixserver.html"> + + + <b>1.4.5.</b> + + Mixing models for a better prediction convergence (MIX server) + + </a> + + + + </li> + + <li class="chapter " data-level="1.4.6" data-path="../tips/emr.html"> + + <a href="../tips/emr.html"> + + + <b>1.4.6.</b> + + Run Hivemall on Amazon Elastic MapReduce + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="1.5" data-path="../tips/general_tips.html"> + + <a href="../tips/general_tips.html"> + + + <b>1.5.</b> + + General Hive/Hadoop Tips + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="1.5.1" data-path="../tips/rowid.html"> + + <a href="../tips/rowid.html"> + + + <b>1.5.1.</b> + + Adding rowid for each row + + </a> + + + + </li> + + <li class="chapter " data-level="1.5.2" data-path="../tips/hadoop_tuning.html"> + + <a href="../tips/hadoop_tuning.html"> + + + <b>1.5.2.</b> + + Hadoop tuning for Hivemall + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="1.6" data-path="../troubleshooting/"> + + <a href="../troubleshooting/"> + + + <b>1.6.</b> + + Troubleshooting + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="1.6.1" data-path="../troubleshooting/oom.html"> + + <a href="../troubleshooting/oom.html"> + + + <b>1.6.1.</b> + + OutOfMemoryError in training + + </a> + + + + </li> + + <li class="chapter " data-level="1.6.2" data-path="../troubleshooting/mapjoin_task_error.html"> + + <a href="../troubleshooting/mapjoin_task_error.html"> + + + <b>1.6.2.</b> + + SemanticException generate map join task error: Cannot serialize object + + </a> + + + + </li> + + <li class="chapter " data-level="1.6.3" data-path="../troubleshooting/asterisk.html"> + + <a href="../troubleshooting/asterisk.html"> + + + <b>1.6.3.</b> + + Asterisk argument for UDTF does not work + + </a> + + + + </li> + + <li class="chapter " data-level="1.6.4" data-path="../troubleshooting/num_mappers.html"> + + <a href="../troubleshooting/num_mappers.html"> + + + <b>1.6.4.</b> + + The number of mappers is less than input splits in Hadoop 2.x + + </a> + + + + </li> + + <li class="chapter " data-level="1.6.5" data-path="../troubleshooting/mapjoin_classcastex.html"> + + <a href="../troubleshooting/mapjoin_classcastex.html"> + + + <b>1.6.5.</b> + + Map-side join causes ClassCastException on Tez + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part II - Generic Features</li> + + + + <li class="chapter " data-level="2.1" data-path="../misc/generic_funcs.html"> + + <a href="../misc/generic_funcs.html"> + + + <b>2.1.</b> + + List of Generic Hivemall Functions + + </a> + + + + </li> + + <li class="chapter " data-level="2.2" data-path="../misc/topk.html"> + + <a href="../misc/topk.html"> + + + <b>2.2.</b> + + Efficient Top-K Query Processing + + </a> + + + + </li> + + <li class="chapter " data-level="2.3" data-path="../misc/tokenizer.html"> + + <a href="../misc/tokenizer.html"> + + + <b>2.3.</b> + + Text Tokenizer + + </a> + + + + </li> + + <li class="chapter " data-level="2.4" data-path="../misc/approx.html"> + + <a href="../misc/approx.html"> + + + <b>2.4.</b> + + Approximate Aggregate Functions + + </a> + + + + </li> + + + + + <li class="header">Part III - Feature Engineering</li> + + + + <li class="chapter " data-level="3.1" data-path="../ft_engineering/scaling.html"> + + <a href="../ft_engineering/scaling.html"> + + + <b>3.1.</b> + + Feature Scaling + + </a> + + + + </li> + + <li class="chapter " data-level="3.2" data-path="../ft_engineering/hashing.html"> + + <a href="../ft_engineering/hashing.html"> + + + <b>3.2.</b> + + Feature Hashing + + </a> + + + + </li> + + <li class="chapter " data-level="3.3" data-path="../ft_engineering/selection.html"> + + <a href="../ft_engineering/selection.html"> + + + <b>3.3.</b> + + Feature Selection + + </a> + + + + </li> + + <li class="chapter " data-level="3.4" data-path="../ft_engineering/binning.html"> + + <a href="../ft_engineering/binning.html"> + + + <b>3.4.</b> + + Feature Binning + + </a> + + + + </li> + + <li class="chapter " data-level="3.5" data-path="../ft_engineering/pairing.html"> + + <a href="../ft_engineering/pairing.html"> + + + <b>3.5.</b> + + Feature Paring + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="3.5.1" data-path="../ft_engineering/polynomial.html"> + + <a href="../ft_engineering/polynomial.html"> + + + <b>3.5.1.</b> + + Polynomial features + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="3.6" data-path="../ft_engineering/ft_trans.html"> + + <a href="../ft_engineering/ft_trans.html"> + + + <b>3.6.</b> + + Feature Transformation + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="3.6.1" data-path="../ft_engineering/vectorization.html"> + + <a href="../ft_engineering/vectorization.html"> + + + <b>3.6.1.</b> + + Feature vectorization + + </a> + + + + </li> + + <li class="chapter " data-level="3.6.2" data-path="../ft_engineering/quantify.html"> + + <a href="../ft_engineering/quantify.html"> + + + <b>3.6.2.</b> + + Quantify non-number features + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="3.7" data-path="../ft_engineering/tfidf.html"> + + <a href="../ft_engineering/tfidf.html"> + + + <b>3.7.</b> + + TF-IDF Calculation + + </a> + + + + </li> + + + + + <li class="header">Part IV - Evaluation</li> + + + + <li class="chapter " data-level="4.1" data-path="../eval/binary_classification_measures.html"> + + <a href="../eval/binary_classification_measures.html"> + + + <b>4.1.</b> + + Binary Classification Metrics + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="4.1.1" data-path="../eval/auc.html"> + + <a href="../eval/auc.html"> + + + <b>4.1.1.</b> + + Area under the ROC curve + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="4.2" data-path="../eval/multilabel_classification_measures.html"> + + <a href="../eval/multilabel_classification_measures.html"> + + + <b>4.2.</b> + + Multi-label Classification Metrics + + </a> + + + + </li> + + <li class="chapter " data-level="4.3" data-path="../eval/regression.html"> + + <a href="../eval/regression.html"> + + + <b>4.3.</b> + + Regression Metrics + + </a> + + + + </li> + + <li class="chapter " data-level="4.4" data-path="../eval/rank.html"> + + <a href="../eval/rank.html"> + + + <b>4.4.</b> + + Ranking Measures + + </a> + + + + </li> + + <li class="chapter " data-level="4.5" data-path="../eval/datagen.html"> + + <a href="../eval/datagen.html"> + + + <b>4.5.</b> + + Data Generation + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="4.5.1" data-path="../eval/lr_datagen.html"> + + <a href="../eval/lr_datagen.html"> + + + <b>4.5.1.</b> + + Logistic Regression data generation + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part V - Supervised Learning</li> + + + + <li class="chapter " data-level="5.1" data-path="../supervised_learning/prediction.html"> + + <a href="../supervised_learning/prediction.html"> + + + <b>5.1.</b> + + How Prediction Works + + </a> + + + + </li> + + <li class="chapter " data-level="5.2" data-path="../supervised_learning/tutorial.html"> + + <a href="../supervised_learning/tutorial.html"> + + + <b>5.2.</b> + + Step-by-Step Tutorial on Supervised Learning + + </a> + + + + </li> + + + + + <li class="header">Part VI - Binary Classification</li> + + + + <li class="chapter " data-level="6.1" data-path="general.html"> + + <a href="general.html"> + + + <b>6.1.</b> + + Binary Classification + + </a> + + + + </li> + + <li class="chapter " data-level="6.2" data-path="a9a.html"> + + <a href="a9a.html"> + + + <b>6.2.</b> + + a9a Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.2.1" data-path="a9a_dataset.html"> + + <a href="a9a_dataset.html"> + + + <b>6.2.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="6.2.2" data-path="a9a_lr.html"> + + <a href="a9a_lr.html"> + + + <b>6.2.2.</b> + + Logistic Regression + + </a> + + + + </li> + + <li class="chapter " data-level="6.2.3" data-path="a9a_minibatch.html"> + + <a href="a9a_minibatch.html"> + + + <b>6.2.3.</b> + + Mini-batch gradient descent + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="6.3" data-path="news20.html"> + + <a href="news20.html"> + + + <b>6.3.</b> + + News20 Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.3.1" data-path="news20_dataset.html"> + + <a href="news20_dataset.html"> + + + <b>6.3.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="6.3.2" data-path="news20_pa.html"> + + <a href="news20_pa.html"> + + + <b>6.3.2.</b> + + Perceptron, Passive Aggressive + + </a> + + + + </li> + + <li class="chapter " data-level="6.3.3" data-path="news20_scw.html"> + + <a href="news20_scw.html"> + + + <b>6.3.3.</b> + + CW, AROW, SCW + + </a> + + + + </li> + + <li class="chapter " data-level="6.3.4" data-path="news20_adagrad.html"> + + <a href="news20_adagrad.html"> + + + <b>6.3.4.</b> + + AdaGradRDA, AdaGrad, AdaDelta + + </a> + + + + </li> + + <li class="chapter " data-level="6.3.5" data-path="news20_rf.html"> + + <a href="news20_rf.html"> + + + <b>6.3.5.</b> + + Random Forest + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="6.4" data-path="kdd2010a.html"> + + <a href="kdd2010a.html"> + + + <b>6.4.</b> + + KDD2010a Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.4.1" data-path="kdd2010a_dataset.html"> + + <a href="kdd2010a_dataset.html"> + + + <b>6.4.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="6.4.2" data-path="kdd2010a_scw.html"> + + <a href="kdd2010a_scw.html"> + + + <b>6.4.2.</b> + + PA, CW, AROW, SCW + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="6.5" data-path="kdd2010b.html"> + + <a href="kdd2010b.html"> + + + <b>6.5.</b> + + KDD2010b Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.5.1" data-path="kdd2010b_dataset.html"> + + <a href="kdd2010b_dataset.html"> + + + <b>6.5.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="6.5.2" data-path="kdd2010b_arow.html"> + + <a href="kdd2010b_arow.html"> + + + <b>6.5.2.</b> + + AROW + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="6.6" data-path="webspam.html"> + + <a href="webspam.html"> + + + <b>6.6.</b> + + Webspam Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.6.1" data-path="webspam_dataset.html"> + + <a href="webspam_dataset.html"> + + + <b>6.6.1.</b> + + Data pareparation + + </a> + + + + </li> + + <li class="chapter " data-level="6.6.2" data-path="webspam_scw.html"> + + <a href="webspam_scw.html"> + + + <b>6.6.2.</b> + + PA1, AROW, SCW + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="6.7" data-path="titanic_rf.html"> + + <a href="titanic_rf.html"> + + + <b>6.7.</b> + + Kaggle Titanic Tutorial + + </a> + + + + </li> + + <li class="chapter " data-level="6.8" data-path="criteo.html"> + + <a href="criteo.html"> + + + <b>6.8.</b> + + Criteo Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="6.8.1" data-path="criteo_dataset.html"> + + <a href="criteo_dataset.html"> + + + <b>6.8.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter active" data-level="6.8.2" data-path="criteo_ffm.html"> + + <a href="criteo_ffm.html"> + + + <b>6.8.2.</b> + + Field-Aware Factorization Machines + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part VII - Multiclass Classification</li> + + + + <li class="chapter " data-level="7.1" data-path="../multiclass/news20.html"> + + <a href="../multiclass/news20.html"> + + + <b>7.1.</b> + + News20 Multiclass Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="7.1.1" data-path="../multiclass/news20_dataset.html"> + + <a href="../multiclass/news20_dataset.html"> + + + <b>7.1.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="7.1.2" data-path="../multiclass/news20_one-vs-the-rest_dataset.html"> + + <a href="../multiclass/news20_one-vs-the-rest_dataset.html"> + + + <b>7.1.2.</b> + + Data preparation for one-vs-the-rest classifiers + + </a> + + + + </li> + + <li class="chapter " data-level="7.1.3" data-path="../multiclass/news20_pa.html"> + + <a href="../multiclass/news20_pa.html"> + + + <b>7.1.3.</b> + + PA + + </a> + + + + </li> + + <li class="chapter " data-level="7.1.4" data-path="../multiclass/news20_scw.html"> + + <a href="../multiclass/news20_scw.html"> + + + <b>7.1.4.</b> + + CW, AROW, SCW + + </a> + + + + </li> + + <li class="chapter " data-level="7.1.5" data-path="../multiclass/news20_ensemble.html"> + + <a href="../multiclass/news20_ensemble.html"> + + + <b>7.1.5.</b> + + Ensemble learning + + </a> + + + + </li> + + <li class="chapter " data-level="7.1.6" data-path="../multiclass/news20_one-vs-the-rest.html"> + + <a href="../multiclass/news20_one-vs-the-rest.html"> + + + <b>7.1.6.</b> + + one-vs-the-rest classifier + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="7.2" data-path="../multiclass/iris.html"> + + <a href="../multiclass/iris.html"> + + + <b>7.2.</b> + + Iris Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="7.2.1" data-path="../multiclass/iris_dataset.html"> + + <a href="../multiclass/iris_dataset.html"> + + + <b>7.2.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="7.2.2" data-path="../multiclass/iris_scw.html"> + + <a href="../multiclass/iris_scw.html"> + + + <b>7.2.2.</b> + + SCW + + </a> + + + + </li> + + <li class="chapter " data-level="7.2.3" data-path="../multiclass/iris_randomforest.html"> + + <a href="../multiclass/iris_randomforest.html"> + + + <b>7.2.3.</b> + + Random Forest + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part VIII - Regression</li> + + + + <li class="chapter " data-level="8.1" data-path="../regression/general.html"> + + <a href="../regression/general.html"> + + + <b>8.1.</b> + + Regression + + </a> + + + + </li> + + <li class="chapter " data-level="8.2" data-path="../regression/e2006.html"> + + <a href="../regression/e2006.html"> + + + <b>8.2.</b> + + E2006-tfidf Regression Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="8.2.1" data-path="../regression/e2006_dataset.html"> + + <a href="../regression/e2006_dataset.html"> + + + <b>8.2.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="8.2.2" data-path="../regression/e2006_arow.html"> + + <a href="../regression/e2006_arow.html"> + + + <b>8.2.2.</b> + + Passive Aggressive, AROW + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="8.3" data-path="../regression/kddcup12tr2.html"> + + <a href="../regression/kddcup12tr2.html"> + + + <b>8.3.</b> + + KDDCup 2012 Track 2 CTR Prediction Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="8.3.1" data-path="../regression/kddcup12tr2_dataset.html"> + + <a href="../regression/kddcup12tr2_dataset.html"> + + + <b>8.3.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="8.3.2" data-path="../regression/kddcup12tr2_lr.html"> + + <a href="../regression/kddcup12tr2_lr.html"> + + + <b>8.3.2.</b> + + Logistic Regression, Passive Aggressive + + </a> + + + + </li> + + <li class="chapter " data-level="8.3.3" data-path="../regression/kddcup12tr2_lr_amplify.html"> + + <a href="../regression/kddcup12tr2_lr_amplify.html"> + + + <b>8.3.3.</b> + + Logistic Regression with amplifier + + </a> + + + + </li> + + <li class="chapter " data-level="8.3.4" data-path="../regression/kddcup12tr2_adagrad.html"> + + <a href="../regression/kddcup12tr2_adagrad.html"> + + + <b>8.3.4.</b> + + AdaGrad, AdaDelta + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part IX - Recommendation</li> + + + + <li class="chapter " data-level="9.1" data-path="../recommend/cf.html"> + + <a href="../recommend/cf.html"> + + + <b>9.1.</b> + + Collaborative Filtering + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="9.1.1" data-path="../recommend/item_based_cf.html"> + + <a href="../recommend/item_based_cf.html"> + + + <b>9.1.1.</b> + + Item-based collaborative filtering + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="9.2" data-path="../recommend/news20.html"> + + <a href="../recommend/news20.html"> + + + <b>9.2.</b> + + News20 Related Article Recommendation Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="9.2.1" data-path="../multiclass/news20_dataset.html"> + + <a href="../multiclass/news20_dataset.html"> + + + <b>9.2.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="9.2.2" data-path="../recommend/news20_jaccard.html"> + + <a href="../recommend/news20_jaccard.html"> + + + <b>9.2.2.</b> + + LSH/MinHash and Jaccard similarity + + </a> + + + + </li> + + <li class="chapter " data-level="9.2.3" data-path="../recommend/news20_knn.html"> + + <a href="../recommend/news20_knn.html"> + + + <b>9.2.3.</b> + + LSH/MinHash and brute-force search + + </a> + + + + </li> + + <li class="chapter " data-level="9.2.4" data-path="../recommend/news20_bbit_minhash.html"> + + <a href="../recommend/news20_bbit_minhash.html"> + + + <b>9.2.4.</b> + + kNN search using b-Bits MinHash + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="9.3" data-path="../recommend/movielens.html"> + + <a href="../recommend/movielens.html"> + + + <b>9.3.</b> + + MovieLens Movie Recommendation Tutorial + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="9.3.1" data-path="../recommend/movielens_dataset.html"> + + <a href="../recommend/movielens_dataset.html"> + + + <b>9.3.1.</b> + + Data preparation + + </a> + + + + </li> + + <li class="chapter " data-level="9.3.2" data-path="../recommend/movielens_cf.html"> + + <a href="../recommend/movielens_cf.html"> + + + <b>9.3.2.</b> + + Item-based collaborative filtering + + </a> + + + + </li> + + <li class="chapter " data-level="9.3.3" data-path="../recommend/movielens_mf.html"> + + <a href="../recommend/movielens_mf.html"> + + + <b>9.3.3.</b> + + Matrix Factorization + + </a> + + + + </li> + + <li class="chapter " data-level="9.3.4" data-path="../recommend/movielens_fm.html"> + + <a href="../recommend/movielens_fm.html"> + + + <b>9.3.4.</b> + + Factorization Machine + + </a> + + + + </li> + + <li class="chapter " data-level="9.3.5" data-path="../recommend/movielens_slim.html"> + + <a href="../recommend/movielens_slim.html"> + + + <b>9.3.5.</b> + + SLIM for fast top-k recommendation + + </a> + + + + </li> + + <li class="chapter " data-level="9.3.6" data-path="../recommend/movielens_cv.html"> + + <a href="../recommend/movielens_cv.html"> + + + <b>9.3.6.</b> + + 10-fold cross validation (Matrix Factorization) + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part X - Anomaly Detection</li> + + + + <li class="chapter " data-level="10.1" data-path="../anomaly/lof.html"> + + <a href="../anomaly/lof.html"> + + + <b>10.1.</b> + + Outlier Detection using Local Outlier Factor (LOF) + + </a> + + + + </li> + + <li class="chapter " data-level="10.2" data-path="../anomaly/sst.html"> + + <a href="../anomaly/sst.html"> + + + <b>10.2.</b> + + Change-Point Detection using Singular Spectrum Transformation (SST) + + </a> + + + + </li> + + <li class="chapter " data-level="10.3" data-path="../anomaly/changefinder.html"> + + <a href="../anomaly/changefinder.html"> + + + <b>10.3.</b> + + ChangeFinder: Detecting Outlier and Change-Point Simultaneously + + </a> + + + + </li> + + + + + <li class="header">Part XI - Clustering</li> + + + + <li class="chapter " data-level="11.1" data-path="../clustering/lda.html"> + + <a href="../clustering/lda.html"> + + + <b>11.1.</b> + + Latent Dirichlet Allocation + + </a> + + + + </li> + + <li class="chapter " data-level="11.2" data-path="../clustering/plsa.html"> + + <a href="../clustering/plsa.html"> + + + <b>11.2.</b> + + Probabilistic Latent Semantic Analysis + + </a> + + + + </li> + + + + + <li class="header">Part XII - GeoSpatial Functions</li> + + + + <li class="chapter " data-level="12.1" data-path="../geospatial/latlon.html"> + + <a href="../geospatial/latlon.html"> + + + <b>12.1.</b> + + Lat/Lon functions + + </a> + + + + </li> + + + + + <li class="header">Part XIII - Hivemall on Spark</li> + + + + <li class="chapter " data-level="13.1" data-path="../spark/getting_started/"> + + <a href="../spark/getting_started/"> + + + <b>13.1.</b> + + Getting Started + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="13.1.1" data-path="../spark/getting_started/installation.html"> + + <a href="../spark/getting_started/installation.html"> + + + <b>13.1.1.</b> + + Installation + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="13.2" data-path="../spark/binaryclass/"> + + <a href="../spark/binaryclass/"> + + + <b>13.2.</b> + + Binary Classification + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="13.2.1" data-path="../spark/binaryclass/a9a_df.html"> + + <a href="../spark/binaryclass/a9a_df.html"> + + + <b>13.2.1.</b> + + a9a tutorial for DataFrame + + </a> + + + + </li> + + <li class="chapter " data-level="13.2.2" data-path="../spark/binaryclass/a9a_sql.html"> + + <a href="../spark/binaryclass/a9a_sql.html"> + + + <b>13.2.2.</b> + + a9a tutorial for SQL + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="13.3" data-path="../spark/binaryclass/"> + + <a href="../spark/binaryclass/"> + + + <b>13.3.</b> + + Regression + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="13.3.1" data-path="../spark/regression/e2006_df.html"> + + <a href="../spark/regression/e2006_df.html"> + + + <b>13.3.1.</b> + + E2006-tfidf regression tutorial for DataFrame + + </a> + + + + </li> + + <li class="chapter " data-level="13.3.2" data-path="../spark/regression/e2006_sql.html"> + + <a href="../spark/regression/e2006_sql.html"> + + + <b>13.3.2.</b> + + E2006-tfidf regression tutorial for SQL + + </a> + + + + </li> + + + </ul> + + </li> + + <li class="chapter " data-level="13.4" data-path="../spark/misc/misc.html"> + + <a href="../spark/misc/misc.html"> + + + <b>13.4.</b> + + Generic features + + </a> + + + + <ul class="articles"> + + + <li class="chapter " data-level="13.4.1" data-path="../spark/misc/topk_join.html"> + + <a href="../spark/misc/topk_join.html"> + + + <b>13.4.1.</b> + + Top-k join processing + + </a> + + + + </li> + + <li class="chapter " data-level="13.4.2" data-path="../spark/misc/functions.html"> + + <a href="../spark/misc/functions.html"> + + + <b>13.4.2.</b> + + Other utility functions + + </a> + + + + </li> + + + </ul> + + </li> + + + + + <li class="header">Part XIV - Hivemall on Docker</li> + + + + <li class="chapter " data-level="14.1" data-path="../docker/getting_started.html"> + + <a href="../docker/getting_started.html"> + + + <b>14.1.</b> + + Getting Started + + </a> + + + + </li> + + + + + <li class="header">Part XIV - External References</li> + + + + <li class="chapter " data-level="15.1" > + + <a target="_blank" href="https://github.com/daijyc/hivemall/wiki/PigHome";> + + + <b>15.1.</b> + + Hivemall on Apache Pig + + </a> + + + + </li> + + + + + <li class="divider"></li> + + <li> + <a href="https://www.gitbook.com"; target="blank" class="gitbook-link"> + Published with GitBook + </a> + </li> +</ul> + + + </nav> + + + </div> + + <div class="book-body"> + + <div class="body-inner"> + + + +<div class="book-header" role="navigation"> + + + <!-- Title --> + <h1> + <i class="fa fa-circle-o-notch fa-spin"></i> + <a href=".." >Field-Aware Factorization Machines</a> + </h1> +</div> + + + + + <div class="page-wrapper" tabindex="-1" role="main"> + <div class="page-inner"> + +<div id="book-search-results"> + <div class="search-noresults"> + + <section class="normal markdown-section"> + + <!-- + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you under the Apache License, Version 2.0 (the + "License"); you may not use this file except in compliance + with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, + software distributed under the License is distributed on an + "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + KIND, either express or implied. See the License for the + specific language governing permissions and limitations + under the License. +--> +<p><a href="https://dl.acm.org/citation.cfm?id=2959134"; target="_blank">Field-aware factorization machines</a> (FFM) is a factorization model which has been used by the <a href="https://www.kaggle.com/c/criteo-display-ad-challenge/discussion/10555"; target="_blank">#1 solution</a> of the Criteo competition.</p> +<p>This page guides you to try the factorization technique with Hivemall's <code>train_ffm</code> and <code>ffm_predict</code> UDFs.</p> +<!-- toc --><div id="toc" class="toc"> + +<ul> +<li><a href="#preprocess-data-and-convert-into-libffm-format">Preprocess data and convert into LIBFFM format</a></li> +<li><a href="#insert-preprocessed-data-into-tables">Insert preprocessed data into tables</a></li> +<li><a href="#training">Training</a></li> +<li><a href="#prediction-and-evaluation">Prediction and evaluation</a></li> +</ul> + +</div><!-- tocstop --> +<div class="panel panel-primary"><div class="panel-heading"><h3 class="panel-title" id="note"><i class="fa fa-edit"></i> Note</h3></div><div class="panel-body"><p>This feature is supported from Hivemall v0.5.1 or later.</p></div></div> +<h1 id="preprocess-data-and-convert-into-libffm-format">Preprocess data and convert into LIBFFM format</h1> +<p>Since FFM is a relatively complex factor-based model which requires us to spend a significant amount of time for feature engineering, preprocessing data outside of Hive can be a reasonable option.</p> +<p>You can again use the repository <strong><a href="https://github.com/takuti/criteo-ffm"; target="_blank">takuti/criteo-ffm</a></strong> cloned in the <a href="criteo_dataset.html">data preparation guide</a> to preprocess the data as the winning solution did:</p> +<pre><code class="lang-sh"><span class="hljs-built_in">cd</span> criteo-ffm +<span class="hljs-comment"># create the CSV files `tr.csv` and `te.csv`</span> +make preprocess +</code></pre> +<p>Task <code>make preprocess</code> executes some Python scripts which are originally taken from <a href="https://github.com/guestwalk/kaggle-2014-criteo"; target="_blank">guestwalk/kaggle-2014-criteo</a> and <a href="https://github.com/chenhuang-learn/ffm"; target="_blank">chenhuang-learn/ffm</a>.</p> +<p>Eventually, you will obtain the following files in so-called LIBFFM format:</p> +<ul> +<li><code>tr.ffm</code> - Labeled training samples<ul> +<li><code>tr.sp</code> - 80% of the labeled training samples randomly picked from <code>tr.ffm</code></li> +<li><code>va.sp</code> - Remaining 20% of samples for evaluation</li> +</ul> +</li> +<li><code>te.ffm</code> - Unlabeled test samples</li> +</ul> +<pre><code><label> <field1>:<feature1>:<value1> <field2>:<feature2>:<value2> ... +. +. +. +</code></pre><p>See <a href="https://github.com/guestwalk/libffm"; target="_blank">LIBFFM official README</a> for detail.</p> +<p>In order to evaluate the accuracy of prediction at the end of this tutorial, later sections use <code>tr.sp</code> and <code>va.sp</code>.</p> +<h1 id="insert-preprocessed-data-into-tables">Insert preprocessed data into tables</h1> +<p>Create new tables used by the FFM UDFs:</p> +<pre><code class="lang-sh">hadoop fs -put tr.sp /criteo/ffm/train +hadoop fs -put va.sp /criteo/ffm/<span class="hljs-built_in">test</span> +</code></pre> +<pre><code class="lang-sql"><span class="hljs-keyword">use</span> criteo; +</code></pre> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> train_ffm; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">EXTERNAL</span> <span class="hljs-keyword">TABLE</span> train_ffm ( + label <span class="hljs-built_in">int</span>, + <span class="hljs-comment">-- quantitative features</span> + i1 <span class="hljs-keyword">string</span>,i2 <span class="hljs-keyword">string</span>,i3 <span class="hljs-keyword">string</span>,i4 <span class="hljs-keyword">string</span>,i5 <span class="hljs-keyword">string</span>,i6 <span class="hljs-keyword">string</span>,i7 <span class="hljs-keyword">string</span>,i8 <span class="hljs-keyword">string</span>,i9 <span class="hljs-keyword">string</span>,i10 <span class="hljs-keyword">string</span>,i11 <span class="hljs-keyword">string</span>,i12 <span class="hljs-keyword">string</span>,i13 <span class="hljs-keyword">string</span>, + <span class="hljs-comment">-- categorical features</span> + c1 <span class="hljs-keyword">string</span>,c2 <span class="hljs-keyword">string</span>,c3 <span class="hljs-keyword">string</span>,c4 <span class="hljs-keyword">string</span>,c5 <span class="hljs-keyword">string</span>,c6 <span class="hljs-keyword">string</span>,c7 <span class="hljs-keyword">string</span>,c8 <span class="hljs-keyword">string</span>,c9 <span class="hljs-keyword">string</span>,c10 <span class="hljs-keyword">string</span>,c11 <span class="hljs-keyword">string</span>,c12 <span class="hljs-keyword">string</span>,c13 <span class="hljs-keyword">string</span>,c14 <span class="hljs-keyword">string</span>,c15 <span class="hljs-keyword">string</span>,c16 <span class="hljs-keyword">string</span>,c17 <span class="hljs-keyword">string</span>,c18 <span class="hljs-keyword">string</span>,c19 <span class="hljs-keyword">string</span>,c20 <span class="hljs-keyword">string</span>,c21 <span class="hljs-keyword">string</span>,c22 <span class="hljs-keyword">string</span>,c23 <span clas s="hljs-keyword">string</span>,c24 <span class="hljs-keyword">string</span>,c25 <span class="hljs-keyword">string</span>,c26 <span class="hljs-keyword">string</span> +) <span class="hljs-keyword">ROW</span> <span class="hljs-keyword">FORMAT</span> +<span class="hljs-keyword">DELIMITED</span> <span class="hljs-keyword">FIELDS</span> <span class="hljs-keyword">TERMINATED</span> <span class="hljs-keyword">BY</span> <span class="hljs-string">' '</span> +<span class="hljs-keyword">STORED</span> <span class="hljs-keyword">AS</span> TEXTFILE LOCATION <span class="hljs-string">'/criteo/ffm/train'</span>; +</code></pre> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> test_ffm; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">EXTERNAL</span> <span class="hljs-keyword">TABLE</span> test_ffm ( + label <span class="hljs-built_in">int</span>, + <span class="hljs-comment">-- quantitative features</span> + i1 <span class="hljs-keyword">string</span>,i2 <span class="hljs-keyword">string</span>,i3 <span class="hljs-keyword">string</span>,i4 <span class="hljs-keyword">string</span>,i5 <span class="hljs-keyword">string</span>,i6 <span class="hljs-keyword">string</span>,i7 <span class="hljs-keyword">string</span>,i8 <span class="hljs-keyword">string</span>,i9 <span class="hljs-keyword">string</span>,i10 <span class="hljs-keyword">string</span>,i11 <span class="hljs-keyword">string</span>,i12 <span class="hljs-keyword">string</span>,i13 <span class="hljs-keyword">string</span>, + <span class="hljs-comment">-- categorical features</span> + c1 <span class="hljs-keyword">string</span>,c2 <span class="hljs-keyword">string</span>,c3 <span class="hljs-keyword">string</span>,c4 <span class="hljs-keyword">string</span>,c5 <span class="hljs-keyword">string</span>,c6 <span class="hljs-keyword">string</span>,c7 <span class="hljs-keyword">string</span>,c8 <span class="hljs-keyword">string</span>,c9 <span class="hljs-keyword">string</span>,c10 <span class="hljs-keyword">string</span>,c11 <span class="hljs-keyword">string</span>,c12 <span class="hljs-keyword">string</span>,c13 <span class="hljs-keyword">string</span>,c14 <span class="hljs-keyword">string</span>,c15 <span class="hljs-keyword">string</span>,c16 <span class="hljs-keyword">string</span>,c17 <span class="hljs-keyword">string</span>,c18 <span class="hljs-keyword">string</span>,c19 <span class="hljs-keyword">string</span>,c20 <span class="hljs-keyword">string</span>,c21 <span class="hljs-keyword">string</span>,c22 <span class="hljs-keyword">string</span>,c23 <span clas s="hljs-keyword">string</span>,c24 <span class="hljs-keyword">string</span>,c25 <span class="hljs-keyword">string</span>,c26 <span class="hljs-keyword">string</span> +) <span class="hljs-keyword">ROW</span> <span class="hljs-keyword">FORMAT</span> +<span class="hljs-keyword">DELIMITED</span> <span class="hljs-keyword">FIELDS</span> <span class="hljs-keyword">TERMINATED</span> <span class="hljs-keyword">BY</span> <span class="hljs-string">' '</span> +<span class="hljs-keyword">STORED</span> <span class="hljs-keyword">AS</span> TEXTFILE LOCATION <span class="hljs-string">'/criteo/ffm/test'</span>; +</code></pre> +<p>Vectorize the LIBFFM-formatted features with <code>rowid</code>:</p> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> train_vectorized; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">TABLE</span> train_vectorized <span class="hljs-keyword">AS</span> +<span class="hljs-keyword">SELECT</span> + row_number() <span class="hljs-keyword">OVER</span> () <span class="hljs-keyword">AS</span> <span class="hljs-keyword">rowid</span>, + <span class="hljs-built_in">array</span>( + i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, + c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11, c12, c13, c14, c15, c16, c17, c18, c19, c20, c21, c22, c23, c24, c25, c26 + ) <span class="hljs-keyword">AS</span> features, + label +<span class="hljs-keyword">FROM</span> + train_ffm +; +</code></pre> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> test_vectorized; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">TABLE</span> test_vectorized <span class="hljs-keyword">AS</span> +<span class="hljs-keyword">SELECT</span> + row_number() <span class="hljs-keyword">OVER</span> () <span class="hljs-keyword">AS</span> <span class="hljs-keyword">rowid</span>, + <span class="hljs-built_in">array</span>( + i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, + c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11, c12, c13, c14, c15, c16, c17, c18, c19, c20, c21, c22, c23, c24, c25, c26 + ) <span class="hljs-keyword">AS</span> features, + label +<span class="hljs-keyword">FROM</span> + test_ffm +; +</code></pre> +<h1 id="training">Training</h1> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> criteo.ffm_model; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">TABLE</span> criteo.ffm_model ( + model_id <span class="hljs-built_in">int</span>, + i <span class="hljs-built_in">int</span>, + Wi <span class="hljs-built_in">float</span>, + Vi <span class="hljs-built_in">array</span><<span class="hljs-built_in">float</span>> +); +</code></pre> +<pre><code class="lang-sql"><span class="hljs-keyword">INSERT</span> OVERWRITE <span class="hljs-keyword">TABLE</span> criteo.ffm_model +<span class="hljs-keyword">SELECT</span> + train_ffm( + features, + label, + <span class="hljs-string">'-init_v random -max_init_value 0.5 -classification -iterations 15 -factors 4 -eta 0.2 -optimizer adagrad -lambda 0.00002'</span> + ) +<span class="hljs-keyword">FROM</span> ( + <span class="hljs-keyword">SELECT</span> + features, label + <span class="hljs-keyword">FROM</span> + criteo.train_vectorized + CLUSTER <span class="hljs-keyword">BY</span> <span class="hljs-keyword">rand</span>(<span class="hljs-number">1</span>) +) t +; +</code></pre> +<p>The third argument of <code>train_ffm</code> accepts a variety of options:</p> +<pre><code>hive> SELECT train_ffm(array(), 0, '-help'); +usage: train_ffm(array<string> x, double y [, const string options]) - + Returns a prediction model [-alpha <arg>] [-auto_stop] [-beta + <arg>] [-c] [-cv_rate <arg>] [-disable_cv] [-enable_norm] + [-enable_wi] [-eps <arg>] [-eta <arg>] [-eta0 <arg>] [-f <arg>] + [-feature_hashing <arg>] [-help] [-init_v <arg>] [-int_feature] + [-iters <arg>] [-l1 <arg>] [-l2 <arg>] [-lambda0 <arg>] [-lambdaV + <arg>] [-lambdaW0 <arg>] [-lambdaWi <arg>] [-max <arg>] [-maxval + <arg>] [-min <arg>] [-min_init_stddev <arg>] [-no_norm] + [-num_fields <arg>] [-opt <arg>] [-p <arg>] [-power_t <arg>] [-seed + <arg>] [-sigma <arg>] [-t <arg>] [-va_ratio <arg>] [-va_threshold + <arg>] [-w0] + -alpha,--alphaFTRL <arg> Alpha value (learning rate) + of + Follow-The-Regularized-Reade + r [default: 0.2] + -auto_stop,--early_stopping Stop at the iteration that + achieves the best validation + on partial samples [default: + OFF] + -beta,--betaFTRL <arg> Beta value (a learning + smoothing parameter) of + Follow-The-Regularized-Reade + r [default: 1.0] + -c,--classification Act as classification + -cv_rate,--convergence_rate <arg> Threshold to determine + convergence [default: 0.005] + -disable_cv,--disable_cvtest Whether to disable + convergence check [default: + OFF] + -enable_norm,--l2norm Enable instance-wise L2 + normalization + -enable_wi,--linear_term Include linear term + [default: OFF] + -eps <arg> A constant used in the + denominator of AdaGrad + [default: 1.0] + -eta <arg> The initial learning rate + -eta0 <arg> The initial learning rate + [default 0.1] + -f,--factors <arg> The number of the latent + variables [default: 5] + -feature_hashing <arg> The number of bits for + feature hashing in range + [18,31] [default: -1]. No + feature hashing for -1. + -help Show function help + -init_v <arg> Initialization strategy of + matrix V [random, + gaussian](default: 'random' + for regression / 'gaussian' + for classification) + -int_feature,--feature_as_integer Parse a feature as integer + [default: OFF] + -iters,--iterations <arg> The number of iterations + [default: 10] + -l1,--lambda1 <arg> L1 regularization value of + Follow-The-Regularized-Reade + r that controls model + Sparseness [default: 0.001] + -l2,--lambda2 <arg> L2 regularization value of + Follow-The-Regularized-Reade + r [default: 0.0001] + -lambda0,--lambda <arg> The initial lambda value for + regularization [default: + 0.0001] + -lambdaV,--lambda_v <arg> The initial lambda value for + V regularization [default: + 0.0001] + -lambdaW0,--lambda_w0 <arg> The initial lambda value for + W0 regularization [default: + 0.0001] + -lambdaWi,--lambda_wi <arg> The initial lambda value for + Wi regularization [default: + 0.0001] + -max,--max_target <arg> The maximum value of target + variable + -maxval,--max_init_value <arg> The maximum initial value in + the matrix V [default: 0.5] + -min,--min_target <arg> The minimum value of target + variable + -min_init_stddev <arg> The minimum standard + deviation of initial matrix + V [default: 0.1] + -no_norm,--disable_norm Disable instance-wise L2 + normalization + -num_fields <arg> The number of fields + [default: 256] + -opt,--optimizer <arg> Gradient Descent optimizer + [default: ftrl, adagrad, + sgd] + -p,--num_features <arg> The size of feature + dimensions [default: -1] + -power_t <arg> The exponent for inverse + scaling learning rate + [default 0.1] + -seed <arg> Seed value [default: -1 + (random)] + -sigma <arg> The standard deviation for + initializing V [default: + 0.1] + -t,--total_steps <arg> The total number of training + examples + -va_ratio,--validation_ratio <arg> Ratio of training data used + for validation [default: + 0.05f] + -va_threshold,--validation_threshold <arg> Threshold to start + validation. At least N + training examples are used + before validation [default: + 1000] + -w0,--global_bias Whether to include global + bias term w0 [default: OFF] +</code></pre><p>Note that debug log describes the change of cumulative loss over iterations as follows:</p> +<pre><code>Iteration #2 | average loss=0.5407147187026483, current cumulative loss=858.114258581103, previous cumulative loss=1682.1101438997914, change rate=0.48985846040280256, #trainingExamples=1587 +Iteration #3 | average loss=0.5105058761578417, current cumulative loss=810.1728254624949, previous cumulative loss=858.114258581103, change rate=0.05586835626980435, #trainingExamples=1587 +Iteration #4 | average loss=0.49045915570992393, current cumulative loss=778.3586801116493, previous cumulative loss=810.1728254624949, change rate=0.039268344174200345, #trainingExamples=1587 +Iteration #5 | average loss=0.4752751205770395, current cumulative loss=754.2616163557617, previous cumulative loss=778.3586801116493, change rate=0.030958816766109738, #trainingExamples=1587 +Iteration #6 | average loss=0.46308523885164105, current cumulative loss=734.9162740575543, previous cumulative loss=754.2616163557617, change rate=0.02564805351182389, #trainingExamples=1587 +Iteration #7 | average loss=0.4529012395753083, current cumulative loss=718.7542672060143, previous cumulative loss=734.9162740575543, change rate=0.02199163009727323, #trainingExamples=1587 +Iteration #8 | average loss=0.44411358945347845, current cumulative loss=704.8082664626703, previous cumulative loss=718.7542672060143, change rate=0.019403016273636577, #trainingExamples=1587 +Iteration #9 | average loss=0.4363264696377158, current cumulative loss=692.450107315055, previous cumulative loss=704.8082664626703, change rate=0.017534072365012268, #trainingExamples=1587 +Iteration #10 | average loss=0.4292753045556725, current cumulative loss=681.2599083298522, previous cumulative loss=692.450107315055, change rate=0.01616029641267912, #trainingExamples=1587 +Iteration #11 | average loss=0.42277515600757143, current cumulative loss=670.9441725840159, previous cumulative loss=681.2599083298522, change rate=0.015142144165104322, #trainingExamples=1587 +Iteration #12 | average loss=0.416689617663307, current cumulative loss=661.2864232316682, previous cumulative loss=670.9441725840159, change rate=0.014394266687126348, #trainingExamples=1587 +Iteration #13 | average loss=0.4109140194740033, current cumulative loss=652.1205489052433, previous cumulative loss=661.2864232316682, change rate=0.013860672175351585, #trainingExamples=1587 +Iteration #14 | average loss=0.4053667348634373, current cumulative loss=643.317008228275, previous cumulative loss=652.1205489052433, change rate=0.013499866998129951, #trainingExamples=1587 +Iteration #15 | average loss=0.3999840450561501, current cumulative loss=634.7746795041102, previous cumulative loss=643.317008228275, change rate=0.013278568131893133, #trainingExamples=1587 +Performed 15 iterations of 1,587 training examples on memory (thus 23,805 training updates in total) +</code></pre><h1 id="prediction-and-evaluation">Prediction and evaluation</h1> +<pre><code class="lang-sql"><span class="hljs-keyword">DROP</span> <span class="hljs-keyword">TABLE</span> <span class="hljs-keyword">IF</span> <span class="hljs-keyword">EXISTS</span> criteo.test_exploded; +<span class="hljs-keyword">CREATE</span> <span class="hljs-keyword">TABLE</span> criteo.test_exploded <span class="hljs-keyword">AS</span> +<span class="hljs-keyword">SELECT</span> + t1.<span class="hljs-keyword">rowid</span>, + t2.i, + t2.j, + t2.Xi, + t2.Xj +<span class="hljs-keyword">from</span> + criteo.test_vectorized t1 + LATERAL <span class="hljs-keyword">VIEW</span> feature_pairs(t1.features, <span class="hljs-string">'-ffm'</span>) t2 <span class="hljs-keyword">AS</span> i, j, Xi, Xj +; +</code></pre> +<pre><code class="lang-sql">WITH predicted AS ( + <span class="hljs-keyword">SELECT</span> + <span class="hljs-keyword">rowid</span>, + <span class="hljs-keyword">avg</span>(score) <span class="hljs-keyword">AS</span> predicted + <span class="hljs-keyword">FROM</span> ( + <span class="hljs-keyword">SELECT</span> + t1.<span class="hljs-keyword">rowid</span>, + p1.model_id, + sigmoid(ffm_predict(p1.Wi, p1.Vi, p2.Vi, t1.Xi, t1.Xj)) <span class="hljs-keyword">AS</span> score + <span class="hljs-keyword">FROM</span> + criteo.test_exploded t1 + <span class="hljs-keyword">JOIN</span> criteo.ffm_model p1 <span class="hljs-keyword">ON</span> (p1.i = t1.i) <span class="hljs-comment">-- at least p1.i = 0 and t1.i = 0 exists</span> + <span class="hljs-keyword">LEFT</span> <span class="hljs-keyword">OUTER</span> <span class="hljs-keyword">JOIN</span> criteo.ffm_model p2 <span class="hljs-keyword">ON</span> (p2.model_id = p1.model_id <span class="hljs-keyword">and</span> p2.i = t1.j) + <span class="hljs-keyword">WHERE</span> + p1.Wi <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">null</span> <span class="hljs-keyword">OR</span> p2.Vi <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">null</span> + <span class="hljs-keyword">GROUP</span> <span class="hljs-keyword">BY</span> + t1.<span class="hljs-keyword">rowid</span>, p1.model_id + ) t + <span class="hljs-keyword">GROUP</span> <span class="hljs-keyword">BY</span> + <span class="hljs-keyword">rowid</span> +) +<span class="hljs-keyword">SELECT</span> + logloss(t1.predicted, t2.label) +<span class="hljs-keyword">FROM</span> + predicted t1 +<span class="hljs-keyword">JOIN</span> + criteo.test_vectorized t2 + <span class="hljs-keyword">ON</span> t1.<span class="hljs-keyword">rowid</span> = t2.<span class="hljs-keyword">rowid</span> +; +</code></pre> +<blockquote> +<p>0.47276208106423234</p> +</blockquote> +<p><br></p> +<div class="panel panel-primary"><div class="panel-heading"><h3 class="panel-title" id="note"><i class="fa fa-edit"></i> Note</h3></div><div class="panel-body"><p>The accuracy varies depending on the random separation of <code>tr.sp</code> and <code>va.sp</code>.</p></div></div> +<p>Notice that LogLoss around 0.45 is reasonable accuracy compared to the <a href="https://github.com/guestwalk/libffm"; target="_blank">competition leaderboard</a> and output from <a href="https://github.com/guestwalk/libffm"; target="_blank">LIBFFM</a>. +<div id="page-footer" class="localized-footer"><hr><!-- + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you under the Apache License, Version 2.0 (the + "License"); you may not use this file except in compliance + with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, + software distributed under the License is distributed on an + "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + KIND, either express or implied. See the License for the + specific language governing permissions and limitations + under the License. +--> +<p><sub><font color="gray"> +Apache Hivemall is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the Apache Incubator. +</font></sub></p> +</div></p> + + + </section> + + </div> + <div class="search-results"> + <div class="has-results"> + + <h1 class="search-results-title"><span class='search-results-count'></span> results matching "<span class='search-query'></span>"</h1> + <ul class="search-results-list"></ul> + + </div> + <div class="no-results"> + + <h1 class="search-results-title">No results matching "<span class='search-query'></span>"</h1> + + </div> + </div> +</div> + + </div> + </div> + + </div> + + + + + </div> + + <script> + var gitbook = gitbook || []; + gitbook.push(function() { + gitbook.page.hasChanged({"page":{"title":"Field-Aware Factorization Machines","level":"6.8.2","depth":2,"next":{"title":"News20 Multiclass Tutorial","level":"7.1","depth":1,"path":"multiclass/news20.md","ref":"multiclass/news20.md","articles":[{"title":"Data preparation","level":"7.1.1","depth":2,"path":"multiclass/news20_dataset.md","ref":"multiclass/news20_dataset.md","articles":[]},{"title":"Data preparation for one-vs-the-rest classifiers","level":"7.1.2","depth":2,"path":"multiclass/news20_one-vs-the-rest_dataset.md","ref":"multiclass/news20_one-vs-the-rest_dataset.md","articles":[]},{"title":"PA","level":"7.1.3","depth":2,"path":"multiclass/news20_pa.md","ref":"multiclass/news20_pa.md","articles":[]},{"title":"CW, AROW, SCW","level":"7.1.4","depth":2,"path":"multiclass/news20_scw.md","ref":"multiclass/news20_scw.md","articles":[]},{"title":"Ensemble learning","level":"7.1.5","depth":2,"path":"multiclass/news20_ensemble.md","ref":"multiclass/news20_ensemble.md","art icles":[]},{"title":"one-vs-the-rest classifier","level":"7.1.6","depth":2,"path":"multiclass/news20_one-vs-the-rest.md","ref":"multiclass/news20_one-vs-the-rest.md","articles":[]}]},"previous":{"title":"Data preparation","level":"6.8.1","depth":2,"path":"binaryclass/criteo_dataset.md","ref":"binaryclass/criteo_dataset.md","articles":[]},"dir":"ltr"},"config":{"plugins":["theme-api","edit-link","github","splitter","sitemap","etoc","callouts","toggle-chapters","anchorjs","codeblock-filename","expandable-chapters","multipart","codeblock-filename","katex","emphasize","localized-footer"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"emphasize":{},"callouts":{},"etoc":{"h2lb":3,"header":1,"maxdepth":3,"mindepth":1,"notoc":true},"github":{"url":"https://github.com/apache/incubator-hivemall/"},"splitter":{},"search":{},"downloadpdf":{"base":"https://g ithub.com/apache/incubator-hivemall/docs/gitbook","label":"PDF","multilingual":false},"multipart":{},"localized-footer":{"filename":"FOOTER.md","hline":"true"},"lunr":{"maxIndexSize":1000000,"ignoreSpecialCharacters":false},"katex":{},"fontsettings":{"theme":"white","family":"sans","size":2,"font":"sans"},"highlight":{},"codeblock-filename":{},"sitemap":{"hostname":"http://hivemall.incubator.apache.org/"},"theme-api":{"languages":[],"split":false,"theme":"dark"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"edit-link":{"label":"Edit","base":"https://github.com/apache/incubator-hivemall/tree/master/docs/gitbook"},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"anchorjs":{"selector":"h1,h2,h3,*:not(.callout) > h4,h5" },"toggle-chapters":{},"expandable-chapters":{}},"theme":"default","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{},"title":"Hivemall User Manual","links":{"sidebar":{"<i class=\"fa fa-home\"></i> Home":"http://hivemall.incubator.apache.org/"}},"gitbook":"3.x.x","description":"User Manual for Apache Hivemall"},"file":{"path":"binaryclass/criteo_ffm.md","mtime":"2018-08-29T08:55:00.265Z","type":"markdown"},"gitbook":{"version":"3.2.3","time":"2018-08-31T06:54:21.013Z"},"basePath":"..","book":{"language":""}}); + }); + </script> +</div> + + + <script src="../gitbook/gitbook.js"></script> +
<TRUNCATED>
