0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 5 years have passed since last update.

a9a dataの分類精度メモ

Last updated at Posted at 2016-01-27

a9a
https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/a9a

を各種分類器で分類した時の精度評価メモ

a9aはlinear separableでないので、これを使って
FFMの分類性能をまずはみる。

liblinear

parameterはデフォルト

closed testの結果は

  • c=100, 1, 0.001で軽く学習。
  • c=1.0、0.001は大差なし。
  • c=100は、69%前後。
masabon:myliblinear masabon$ ./predict.sh ./data/a9a_train.svmdata a9a_train.svmdata.model o
Accuracy = 84.9237% (27652/32561)

masabon:myliblinear masabon$ ./train.sh data/a4a 
...........*.
optimization finished, #iter = 120
Objective value = -2007.96
nSV = 2922
masabon:myliblinear masabon$ ./predict.sh data/a4a a4a.model o2
Accuracy = 84.9822% (4063/4781)

myffm

  • k = 100
  • maxiter =20
  • #train = 5000 (a4a相当)
  • 学習率は固定。要、adagra, adadelta対応
  • SGD実装
  • closed testなのであくまで参考程度
  • 次はcvで
  • #train = 1000の場合は、90%程度いった。処理が遅いので、maxiterを増やすのにも時間がかかる。要高速化
  • field idは単純に列の順番、feature idはそのまま、valueは1固定
  • linear termはoff設定
  • lambda = 0.0001, alpha=0.01固定
%19----
pp: 871
pn: 367
np: 364
nn: 3398
acc: 0.8538
likelihood=117605.88086645157
%%%
pp: 897
pn: 341
np: 318
nn: 3444
acc: 0.8682

# 44----
pp: 943
pn: 295
np: 288
nn: 3474
acc: 0.8834
likelihood=83245.71565255232
---
pp: 1097
pn: 141
np: 493
nn: 3269
acc: 0.8732
#############

#############
# 49----
pp: 950
pn: 288
np: 270
nn: 3492
acc: 0.8884
likelihood=78719.89344336146
---
pp: 1092
pn: 146
np: 481
nn: 3281
acc: 0.8746

#train=1000の場合

closed testではあるが、90%超になる。
2次の項はそれなりに効いている模様

#############
# 11----
pp: 178
pn: 58
np: 57
nn: 707
acc: 0.885
likelihood=56379.20298554983
---
pp: 187
pn: 49
np: 39
nn: 725
acc: 0.912

.
.
.
# 49----
pp: 221
pn: 15
np: 15
nn: 749
acc: 0.97
likelihood=49211.286605422545
---
pp: 230
pn: 6
np: 55
nn: 709
acc: 0.939
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?