No convergence

[ricardopolo]

Hello.

I am getting this error
node_modules/numeric/numeric-1.2.6.js:4340 throw 'Error: no convergence.'

I passed this data https://raw.githubusercontent.com/mailgun/talon/master/talon/signature/data/train.data
where the last position is the expected outcome.

This is the code that throws the error.

  fs.readFile('traineddata.txt', 'utf8', function(err, content){
    var train_data = [];
    S(content.trim()).lines().forEach(function(line){
      var spllited_line = line.split(',');
      supervisory_signal = spllited_line[spllited_line.length - 1];
      var vector = [];
      for (i = 0; i < spllited_line.length -1; i++) {
        vector.push(spllited_line[i]);
      }
      train_data.push([vector, supervisory_signal]);
    });
      var clf = new svm.SVM();
     clf.train(train_data).done(function () {
    var prediction = clf.predictSync(input);
  });
  });

Any advice about how can I fix it?

[nicolaspanel]

I think you are passing Strings instead of Numbers.

Something like this should work:

fs.readFile('traineddata.txt', 'utf8', function(err, content){
    var data = S(content.trim()).lines().map(function(line){
      line = line.split(',').map(function(item){ return parseFloat(item); });
      return [ line.slice(0,-1)  , line.slice(-1)[0]];
     });
      var clf = new svm.SVM();
     clf.train(data).done(function () {
    var prediction = clf.predictSync(input);
  }); 
});

[mjp0]

I'm having this same issue. I've triple checked that all my numbers are actual numbers.

Here's my trainer code. At this point I'm just trying to get this library to do anything at all so I'm just copy/pasting from examples. Data is fetched from database so I'm not using svm.read() or similar functions.

const clf = new svm.EpsilonSVR({
    gamma: [0.125, 0.5, 1],
    c: [8, 16, 32],
    epsilon: [0.001, 0.125, 0.5],
    normalize: true, // (default)
    reduce: true, // (default)
    retainedVariance: 0.995,
    kFold: 5
});
clf.train(dataset)
        .progress(function(progress){
            console.log('training progress: %d%', Math.round(progress*100));
        })
        .spread(function (model, report) {
            console.log('SVM trained. \nReport :\n%s', so(report));
            return dataset;
        }).then(function (dataset) {
            // randomly pick m values and display predictions
            _a.pick(dataset, 5).forEach(function (ex, i) {
                var prediction = clf.predictSync(ex[0]);
                console.log(' { #%d, expected: %d, predicted: %d}',i+1, ex[1], prediction);
            });
        })
        .fail(function (err) {
            throw err;

        })
        .done(function () {
            console.log('done.');
        });

Here's what my dataset looks like:

[ [ [ 0.9370528277403535,
      1,
      10.2446782417689,
      0.1,
      0.2157508638191752,
      0.4,
      2.9547269278195842,
      0.6,
      5.305598923648802,
      0.7,
      14.080959534461057,
      0.8,
      16.787868319892056,
      0.9,
      9.33018890638979,
      0.2,
      0.6232436952147369,
      0.3,
      0.21013282524728893,
      0.5,
      3.74996578411525,
      0.06294717225964656,
      1,
      6.780877019575415,
      2,
      1.9003776435045472,
      2.1,
      10.242337961671112,
      1.6,
      6.317515476417146,
      1.5,
      12.969169488832977,
      1.3,
      12.570850329693071,
      1.2,
      2.327629391532482,
      0.9,
      5.074429387365065,
      0.7,
      3.3060673703062537,
      0.4,
      3.9453340184839814,
      0.1,
      0.15236271737872897,
      0.2,
      0.2373174670956546,
      0.3,
      0.8227919716677468,
      0.5,
      3.1718293970518285,
      0.6,
      2.89604238058212,
      0.8,
      1.5460014704966607,
      1.1,
      5.111847470474484,
      1.4,
      3.97480070903542,
      1.7,
      6.72969175025937,
      1.8,
      3.49031097535919,
      1.9,
      7.411393336684169 ],
    0 ],
  [ [ 0.9369869025261525,
      1,
      1.819219777880099,
      0.1,
      0.18802582343899754,
      0.3,
      0.8968994041430811,
      0.5,
      4.825678824413293,
      0.6,
      11.915415800114726,
      0.8,
      18.48834287306935,
      0.9,
      12.06725732683406,
      0.2,
      0.3907571402120391,
      0.4,
      2.336214726853436,
      0.7,
      6.502915428451727,
      0.06301309747384755,
      1,
      2.7389554182335907,
      1.7,
      6.555133166011573,
      1.6,
      7.556795774747793,
      1.4,
      9.463475271773756,
      1.2,
      1.1956521739130421,
      1.1,
      12.601448919248993,
      0.9,
      4.353193433215896,
      0.8,
      4.3536001430615645,
      0.5,
      5.420907071538794,
      0.1,
      0.25291109688767577,
      0.2,
      0.17735759446704247,
      0.3,
      1.3192241076551925,
      0.4,
      1.0979702653683976,
      0.6,
      3.265845573077364,
      0.7,
      3.2684995321619725,
      1.3,
      6.354655877994975,
      1.5,
      7.364235879648184,
      1.8,
      8.743032992838561,
      1.9,
      5.596492067848269 ],
    2 ] ]

[nicolaspanel]

Please check your data

[mjp0]

@nicolaspanel check it for what?

[viorelzavoiu]

Don't know if this helps, but i had the same issue and I found out that there was a problem with the reduction part of the algorithm with PCA. The error appears in the Single Value Decomposition procedure. So to bypass the error I skipped the reduction step: { reduce : false }