testsmf.ssc

import BIDMach.models.SMF
val dir="../data/netflix/"

//val a0=loadSMat(dir+"train.smat.lz4")
//val ta0=loadSMat(dir+"test.smat.lz4")
val a0=loadSMat(dir+"newtrain.smat.lz4")
val ta0=loadSMat(dir+"newtest.smat.lz4")

val rebuild = false

val (a, ta) = if (rebuild) {
val (ii, jj, vv) = find3(a0+ta0);
val ir = rand(ii.length, 1) < 0.1f;
val itrain = find(ir == 0);
val itest = find(ir);
val a = sparse(ii(itrain),jj(itrain),vv(itrain),a0.nrows,a0.ncols);
val ta = sparse(ii(itest),jj(itest),vv(itest),a0.nrows,a0.ncols);
saveSMat(dir+"newtrain.smat.lz4", a)
saveSMat(dir+"newtest.smat.lz4", ta)
(a, ta)
} else {
(a0, ta0)
}

val d = 512

val fact = zeros(d, a.ncols);
val (nn,opts) = SMF.learnerX(a,fact,d)

// good values for movielens 10m
opts.lambdam = 5f
opts.lambdau = 5f
opts.uiter = 8
opts.npasses = 5
opts.batchSize = 2000

// good values for netflix
opts.batchSize = 1000
opts.uiter = 2
// for acceptance = 1
opts.urate = 0.01f
opts.lrate = 0.01f  
// for computed acceptance
opts.urate = 0.1f
opts.lrate = 0.1  
opts.npasses = 5

//opts.texp = null
//opts.aopts = null
val lambda = 4f
opts.lambdau = lambda;
opts.regumean = lambda;
opts.lambdam = lambda / 500000 * 20;
opts.regmmean = opts.lambdam
opts.evalStep = 31
opts.doUsers = false
opts.lsgd = 0.010f
//opts.weightByUser = true
//opts.traceConverge = true
//opts.autoReset = false
//opts.useDouble = true

nn.train

val mmat = FMat(nn.modelmat)
val iavg = FMat(nn.modelmats(1))
val avg = FMat(nn.modelmats(2))

val model = nn.model.asInstanceOf[SMF]
val (mm, mopts) = SMF.predictor(model, a, fact);
mopts.uiter = 6
mopts.urate = opts.urate
mopts.lsgd = 0.0f

def testme = {
mm.predict
val mmod = mm.model.asInstanceOf[SMF];
val tanz = ta > 0;
val preds = DDS(FMat(nn.modelmats(0)), fact, tanz) + FMat(nn.modelmats(1) + nn.modelmats(2))
min(preds.contents, 5f, preds.contents)
max(preds.contents, 1f, preds.contents)
val diff = DMat(preds.contents - ta.contents)
val rmse = sqrt((diff ^* diff) / diff.length.toFloat).dv
rmse
}
val rmse = testme

println("rmse = %f" format rmse);