Skip to content

Linkers evaluation

Andrea Tupini edited this page May 9, 2019 · 12 revisions

Setting

  • run: April 11 2019 on soweego-1 VPS instance;
  • output folder: /srv/dev/20190411;
  • head commit: 1505429997b878568a9e24185dc3afa7ad4720eb;
  • command: python -m soweego linker evaluate ${Algorithm} ${Dataset} ${Entity};
  • evaluation technique: stratified 5-fold cross validation over training/test splits;
  • mean performance scores over the folds.

Algorithms parameters

  • Naïve Bayes (NB):
    • binarize = 0.1;
    • alpha = 0.0001;
  • liblinear SVM (LSVM): default parameters as per scikit LinearSVC;
  • libsvm SVM (SVM):
    • kernel = linear;
    • other parameters as per scikit SVC defaults;
  • single-layer perceptron (SLP):
    • layer = fully connected (Dense);
    • activation = sigmoid;
    • optimizer = stochastic gradient descent;
    • loss = binary cross-entropy;
    • training batch size = 1,024;
    • training epochs = 100.
  • multi-layer perceptron (MLP):
    • layers = 128 > BN > 32 > BN > 1
      • fully connected layers followed by BatchNormalization (BN)
    • activation:
      • hidden layers = relu;
      • output layer = sigmoid;
    • optimizer = Adadelta;
    • loss = binary cross-entropy
    • training batch size = 1,024;
    • training epochs = 1000;
    • early stopping:
      • patience = 100;

Performance

Algorithm Dataset Entity Precision (std) Recall (std) F-score (std)
NB Discogs Band .789 (.0031) .941 (.0004) .859 (.002)
LSVM Discogs Band .785 (.0058) .946 (.0029) .858 (.0034)
SVM Discogs Band .777 (.003) .963 (.0016) .86 (.0024)
SLP Discogs Band .776 (.0041) .956 (.0012) .857 (.0029)
NB Discogs Musician .836 (.0018) .958 (.0012) .893 (.0013)
SVM Discogs Musician .814 (.0015) .986 (.0003) .892 (.001)
SLP Discogs Musician .815 (.002) .985 (.0006) .892 (.0012)
NB IMDb Actor TODO TODO TODO
SVM IMDb Actor TODO TODO TODO
SLP IMDb Actor TODO TODO TODO
MLP IMDb Actor TODO TODO TODO
NB IMDb Director .897 (.00195) .971 (.0012) .932 (.001)
SVM IMDb Director .919 (.0031) .942 (.0019) .93 (.002)
SLP IMDb Director .867 (.0115) .953 (.0043) .908 (.0056)
NB IMDb Musician .891 (.0042) .96 (.0022) .924 (.0026)
SVM IMDb Musician .917 (.0043) .937 (.0034) .927 (.003)
SLP IMDb Musician .922 (.005) .914 (.0092) .918 (.0055)
NB IMDb Producer .871 (.0023) .97 (.0037) .918 (.0011)
SVM IMDb Producer .92 (.005) .938 (.0038) .929 (.0026)
SLP IMDb Producer .862 (.0609) .914 (.0648) .883 (.0185)
NB IMDb Writer .91 (.003) .961 (.0022) .935 (.0022)
SVM IMDb Writer .936 (.0029) .948 (.0025) .942 (.0026)
SLP IMDb Writer .903 (.0154) .955 (.0147) .928 (.0047)
NB MusicBrainz Band .822 (.00169) .985 (.0008) .896 (.001)
SVM MusicBrainz Band .943 (.0019) .888 (.0027) .914 (.0016)
SLP MusicBrainz Band .93 (.0265) .885 (.0103) .907 (.0082)
NB MusicBrainz Musician .955 (.0009) .936 (.0011) .946 (.00068)
SVM MusicBrainz Musician .941 (.0011) .962 (.001) .952 (.0004)
SLP MusicBrainz Musician .943 (.0018) .956 (.0019) .949 (.0007)

Confidence

The following plots display the confidence scores distribution and the total predictions yielded by each algorithm on each target classification set.

Note that linear SVM is omitted since it does not output probability scores.

Axes:

  • x = # predictions;
  • y = confidence score.

Discogs band

NB, SVM, SLP. MLP

Discogs musician

NB, SVM, SLP. MLP

IMDb director

NB, SVM, SLP. MLP

IMDb musician

NB, SVM, SLP. MLP

IMDb producer

NB, SVM, SLP. MLP

IMDb writer

NB, SVM, SLP. MLP

MusicBrainz band

NB, SVM, SLP. MLP

MusicBrainz musician

NB, SVM, SLP. MLP

Comparison

See the plots above to have a rough idea on the amount of confident predictions.

Threshold values:

  • # predictions >= 0.0000000001, i.e., equivalent to almost all matches;
  • # confident >= 0.8.

Discogs band

WD items: 50,316

Measure NB LSVM SVM SLP MLP
Precision .789 .785 .777 .776 .833
Recall .941 .946 .963 .957 .914
F-score .859 .858 .86 .857 .872
# predictions 820 51 94,430 91,295 91,132
# confident 219 N.A. 1,660 5,355 11,114

Discogs musician

WD items: 199,180

Measure NB LSVM SVM SLP MLP
Precision .836 .814 .815 .815 .849
Recall .958 .986 .985 .985 .961
F-score .893 .892 .892 .892 .902
# predictions 3,872 200 533,301 517,450 514,488
# confident 1,101 N.A. 98,172 58,437 57,184

IMDb director

WD items: 9,249

Measure NB LSVM SVM SLP MLP
Precision .897 .919 .908 .867 .916
Recall .971 .942 .958 .953 .961
F-score .932 .93 .932 .908 .938
# predictions 192 10 17,557 17,187 16,881
# confident 60 N.A. 1,616 553 1,810

IMDb musician

WD items: 217,139

Measure NB LSVM SVM SLP MLP
Precision .891 .917 .908 .922 .903
Recall .96 .937 .942 .914 .951
F-score .924 .927 .924 .918 .926
# predictions 4,806 218 406,674 398,346 376,857
# confident 1,341 N.A. 21,462 7,244 16,272

IMDb producer

WD items: 2,251

Measure NB LSVM SVM SLP MLP
Precision .871 .92 .923 .862 .912
Recall .97 .938 .926 .914 .956
F-score .918 .929 .925 .883 .933
# predictions 56 3 5,249 5,116 5,094
# confident 15 N.A. 507 180 529

IMDb writer

WD items: 16,446

Measure NB LSVM SVM SLP MLP
Precision .91 .936 .932 .903 .921
Recall .961 .948 .954 .955 .962
F-score .935 .942 .943 .928 .941
# predictions 428 17 45,122 44,338 43,868
# confident 138 N.A. 2,934 1,548 3,234

MusicBrainz band

WD items: 32,658

Measure NB LSVM SVM SLP MLP
Precision .822 .943 .939 .93 .933
Recall .985 .888 .893 .885 .902
F-score .896 .914 .915 .907 .918
# predictions 265 33 39,618 38,012 33,981
# confident 46 N.A. 1,475 501 1,506

MusicBrainz musician

WD items: 153,725

Measure NB LSVM SVM SLP MLP
Precision .955 .941 .95 .943 .940
Recall .936 .962 .938 .956 .968
F-score .946 .952 .944 .949 .954
# predictions 2,833 154 280,029 260,530 194,505
# confident 1,212 N.A. 7,496 7,339 8,470