libfm的基本使用
慕铭
参考文献：http://blog.csdn.net/chloezhao/article/details/53462411
话说fm、ffm等为代表的算法在ctr预测方面已经是大显身手，fm、ffm的开源实现台湾国立大学，业界常用的LibSVM， Liblinear等都是他们开发的，开源代码的效率和质量都非常高，本篇文章也是为了介绍libfm的基本使用，libfm可以用于推荐系统，也可以在计算广告行业用于ctr预测。
下载编译：
http://www.libfm.org/convert libFM transpose
convert︰ 将文字文件转换成二进制格式的工具
libFM: 用于训练模型的工具
transpos︰ 转换二进制设计矩阵的一个工具
参数情况：
先运行./libFM -help看看其自带的参数:
----------------------------------------------------------------------------
libFM
Version: 1.4.2
Author: Steffen Rendle, srendle@libfm.org
WWW: http://www.libfm.org/
This program comes with ABSOLUTELY NO WARRANTY; for details see license.txt.
This is free software, and you are welcome to redistribute it under certain
conditions; for details see license.txt.
----------------------------------------------------------------------------
-cache_size cache size for data storage (only applicable if data is
in binary format), default=infty
-dim 'k0,k1,k2': k0=use bias, k1=use 1-way interactions,
k2=dim of 2-way interactions; default=1,1,8
-help this screen
-init_stdev stdev for initialization of 2-way factors; default=0.1
-iter number of iterations; default=100
-learn_rate learn_rate for SGD; default=0.1
-meta filename for meta information about data set
-method learning method (SGD, SGDA, ALS, MCMC); default=MCMC
-out filename for output
-regular 'r0,r1,r2' for SGD and ALS: r0=bias regularization,
r1=1-way regularization, r2=2-way regularization
-relation BS: filenames for the relations, default=''
-rlog write measurements within iterations to a file;
default=''
-task r=regression, c=binary classification [MANDATORY]
-test filename for test data [MANDATORY]
-train filename for training data [MANDATORY]
-validation filename for validation data (only for SGDA)
-verbosity how much infos to print; default=0可以看到-task参数可以选择是r 和 c ，其中r可以用于回归，c可以用于分类，r可以用于推荐系统中等场景，
选择c可以用于ctr预测等场景，-iter表示迭代次数，默认为100次，-method表示优化的方法，可以用sgb，也可以用ALS，好像spark里面自带一个推荐系统的算法就是用的ALS，默认是MCMC这个优化方法没有任何正则选项， -learn_rate是学习率，这个参数一定大于0，一般小于1，-train表示训练的文本文件，这个参数对数据格式要求有两种一种是libsvm格式文件，另外一种是二进制文件(说实话我也不知道这是什么格式文件，可以参考我所参考的wen'xian)
-out选项用于测试文件的预测结果，
我用我的数据做下文本分类，训练数据和测试数据是之前转化好libsvm格式文件:
./libFM -task r -train train.txt -test test.txt -dim ’1,1,8’ -iter 100 -method sgd -learn_rate 0.01 -regular ’0,0,0.01’ -out model.txt训练过程如下:
----------------------------------------------------------------------------
libFM
Version: 1.4.2
Author: Steffen Rendle, srendle@libfm.org
WWW: http://www.libfm.org/
This program comes with ABSOLUTELY NO WARRANTY; for details see license.txt.
This is free software, and you are welcome to redistribute it under certain
conditions; for details see license.txt.
----------------------------------------------------------------------------
Loading train...
has x = 1
has xt = 0
num_rows=61576 num_values=3998509 num_features=209260 min_target=0 max_target=1
Loading test...
has x = 1
has xt = 0
num_rows=15436 num_values=1005321 num_features=209258 min_target=0 max_target=1
#relations: 0
Loading meta data...
learnrate=0.01
learnrates=0.01,0.01,0.01
#iterations=100
SGD: DON'T FORGET TO SHUFFLE THE ROWS IN TRAINING DATA TO GET THE BEST RESULTS.
#Iter= 0 Train=0.347366 Test=0.350806
#Iter= 1 Train=0.288766 Test=0.294584
#Iter= 2 Train=0.260503 Test=0.268643
#Iter= 3 Train=0.243147 Test=0.253436
#Iter= 4 Train=0.231104 Test=0.243393
#Iter= 5 Train=0.22209 Test=0.236236
#Iter= 6 Train=0.214967 Test=0.230852
#Iter= 7 Train=0.209114 Test=0.226627
#Iter= 8 Train=0.204162 Test=0.22322
#Iter= 9 Train=0.199886 Test=0.220407
#Iter= 10 Train=0.196128 Test=0.218044
#Iter= 11 Train=0.192778 Test=0.216032
#Iter= 12 Train=0.189756 Test=0.214291
#Iter= 13 Train=0.187004 Test=0.212767
#Iter= 14 Train=0.184479 Test=0.211426
#Iter= 15 Train=0.182146 Test=0.210239
#Iter= 16 Train=0.179977 Test=0.209181
#Iter= 17 Train=0.177952 Test=0.208231
#Iter= 18 Train=0.176051 Test=0.20737
#Iter= 19 Train=0.17426 Test=0.206581
#Iter= 20 Train=0.172567 Test=0.205863
#Iter= 21 Train=0.170962 Test=0.205207
#Iter= 22 Train=0.169436 Test=0.204604
#Iter= 23 Train=0.167981 Test=0.204051
#Iter= 24 Train=0.166591 Test=0.203541
#Iter= 25 Train=0.165259 Test=0.203069
#Iter= 26 Train=0.163982 Test=0.202631
#Iter= 27 Train=0.162755 Test=0.202224
#Iter= 28 Train=0.161574 Test=0.201845
#Iter= 29 Train=0.160435 Test=0.201489
#Iter= 30 Train=0.159336 Test=0.201155
#Iter= 31 Train=0.158274 Test=0.200843
#Iter= 32 Train=0.157246 Test=0.200546
#Iter= 33 Train=0.15625 Test=0.200266
#Iter= 34 Train=0.155285 Test=0.200002
#Iter= 35 Train=0.154348 Test=0.199752
#Iter= 36 Train=0.153438 Test=0.199518
#Iter= 37 Train=0.152553 Test=0.199298
#Iter= 38 Train=0.151692 Test=0.199091
#Iter= 39 Train=0.150851 Test=0.198896
#Iter= 40 Train=0.150031 Test=0.198712
#Iter= 41 Train=0.149231 Test=0.198539
#Iter= 42 Train=0.148449 Test=0.198375
#Iter= 43 Train=0.147683 Test=0.198219
#Iter= 44 Train=0.146933 Test=0.198071
#Iter= 45 Train=0.1462 Test=0.197931
#Iter= 46 Train=0.145483 Test=0.197799
#Iter= 47 Train=0.144781 Test=0.197673
#Iter= 48 Train=0.144094 Test=0.197553
#Iter= 49 Train=0.143421 Test=0.197441
#Iter= 50 Train=0.142761 Test=0.197334
#Iter= 51 Train=0.142114 Test=0.197233
#Iter= 52 Train=0.14148 Test=0.197138
#Iter= 53 Train=0.140855 Test=0.197044
#Iter= 54 Train=0.140241 Test=0.196956
#Iter= 55 Train=0.139637 Test=0.196871
#Iter= 56 Train=0.139043 Test=0.19679
#Iter= 57 Train=0.138459 Test=0.196714
#Iter= 58 Train=0.137885 Test=0.196642
#Iter= 59 Train=0.13732 Test=0.196574
#Iter= 60 Train=0.136765 Test=0.196509
#Iter= 61 Train=0.136218 Test=0.196448
#Iter= 62 Train=0.135678 Test=0.196391
#Iter= 63 Train=0.135147 Test=0.196337
#Iter= 64 Train=0.134623 Test=0.196285
#Iter= 65 Train=0.134107 Test=0.196236
#Iter= 66 Train=0.133598 Test=0.196189
#Iter= 67 Train=0.133096 Test=0.196144
#Iter= 68 Train=0.132601 Test=0.196103
#Iter= 69 Train=0.132113 Test=0.196064
#Iter= 70 Train=0.131631 Test=0.196027
#Iter= 71 Train=0.131156 Test=0.195992
#Iter= 72 Train=0.130687 Test=0.195959
#Iter= 73 Train=0.130225 Test=0.195928
#Iter= 74 Train=0.129768 Test=0.195899
#Iter= 75 Train=0.129316 Test=0.195871
#Iter= 76 Train=0.12887 Test=0.195844
#Iter= 77 Train=0.12843 Test=0.195819
#Iter= 78 Train=0.127995 Test=0.195797
#Iter= 79 Train=0.127566 Test=0.195775
#Iter= 80 Train=0.127141 Test=0.195756
#Iter= 81 Train=0.126722 Test=0.195737
#Iter= 82 Train=0.126308 Test=0.195721
#Iter= 83 Train=0.125899 Test=0.195705
#Iter= 84 Train=0.125495 Test=0.19569
#Iter= 85 Train=0.125095 Test=0.195676
#Iter= 86 Train=0.124699 Test=0.195664
#Iter= 87 Train=0.124308 Test=0.195653
#Iter= 88 Train=0.123922 Test=0.195643
#Iter= 89 Train=0.123539 Test=0.195634
#Iter= 90 Train=0.123161 Test=0.195624
#Iter= 91 Train=0.122786 Test=0.195616
#Iter= 92 Train=0.122415 Test=0.195609
#Iter= 93 Train=0.122047 Test=0.195603
#Iter= 94 Train=0.121683 Test=0.195598
#Iter= 95 Train=0.121323 Test=0.195595
#Iter= 96 Train=0.120965 Test=0.195592
#Iter= 97 Train=0.120611 Test=0.19559
#Iter= 98 Train=0.120259 Test=0.19559
#Iter= 99 Train=0.119911 Test=0.19559
Final Train=0.119911 Test=0.19559
看下模型预测结果文件:
0
1
1
0
1
1
0
0.260587
0
1
1
0.717071
0
1
预测出来的是一个概率
没有完善，后面继续更新libffm，
类似资料：