public void getModels(int len, double error, ref int length)
{
length = 0;
int kk = (int)Math.Floor((double)ts.Length / len);
models = new Model[kk];
int x = 0;
this.len = len;
for (int i = 0; i < kk; i++)
{
if (i == kk - 1) len = ts.Length - x;
int s = x;
double[] u = new double[len];
for (int j = 0; j < len; j++) u[j] = ts.data[x++];
TimeSeries t = new TimeSeries(u, ts.freq);
Model m = new Model(t);
length += m.countError(error);
m.start = s;
models[i] = m;
}
}
static ArrayList Level(ArrayList ranges, double errror_level)
{
ArrayList tss = null;
ArrayList R = new ArrayList();
foreach (Range r in ranges)
R.Add(new Range(r.s, r.e));
ArrayList X = ranges;
// should ont use computed and matched
for (; ; )
{
int count = 0;
tss = ts.Divide(X);
foreach (TimeSeries t in tss)
{
if (t.r.computed == 1) {if (t.r.matched== 1) count++; continue; }
t.r.computed = 1;
t.r.matched = 0;
Model m;
if(Global.quick)
m=Model.ModelQuick(t);
else
m = new Model(t);//
if (m.error < errror_level)
{
t.r.matched = 1;
bool ignore = false; // may be also be called as do not add to the
count++;
//remove ranges
ArrayList todel = new ArrayList();
foreach (Range r in R)
{
if ((t.r.overlap(r)) && ((int)t.r.len != (int)r.len)) todel.Add(r);
else if (t.r.overlap(r)) ignore = true;
}
foreach (Range r in todel)
R.Remove(r);
if(ignore==false) R.Add(t.r);
}
}
if (count == 0) break;
//if (count == 1) break;
X.Sort();
X = combine(X);
// we need to deset matched or not
}
return R;
}
public void Solve()
{
if (type == ModelType.PLA)
{
values = new double[2];
double asum = 0;
double bsum = 0;
for (int i = 0; i < this.ts.Length; i++)
{
asum += ((i + 1) - ((ts.Length + 1) / 2)) * ts.data[i];
bsum += ((i + 1) - ((2 * ts.Length + 1) / 3)) * ts.data[i];
}
values[0] = 12 * asum / ts.Length / (ts.Length + 1) / (ts.Length - 1);
values[1] = 6 * bsum / ts.Length / (1 - ts.Length);
return;
}
int n = ts.Length;
int l = 0;
freq = ts.freq[0];
while (freq > ts.Length)
{
if (l == ts.freq.Length)
{
return;
}
freq = ts.freq[l++];
}
if (freq == 0)
{
//use regression
values = ChebyshevReg.Solve(ts.data);
seasonal = null;
}
else
{
double[] season_ = new double[freq];
decompose(n, freq, season_);
seasonal = new Model();
int[] f;
if (ts.freq.Length == 1) f = null;
else
{
f = new int[ts.freq.Length - 1];
for (int i = 0; i < ts.freq.Length - 1; i++) f[i] = ts.freq[i + 1];
}
seasonal.ts = new TimeSeries(season_, f);
if (f != null)
seasonal.Solve();
}
}
public static Model ModelQuick(TimeSeries ts)
{
Model m = new Model();
m.ts = ts;
m.seasonal = null;
m.values = null;
m.errors = null;
m.type = ModelType.Explicit;
m.Solve();
m.CalcError();
// m.ComputeErrorRange();
m.error = m.Error(Global.confidence);
m.len = ts.Length;
return m;
}
public Model()
{
ts = null;
seasonal = null;
values = null;
errors = null;
}
public Model(TimeSeries ts)
{
this.ts = ts;
seasonal = null;
values = null;
errors = null;
ModelType best = ModelType.Trend;
double min = double.MaxValue;
foreach (ModelType t in Enum.GetValues(typeof(ModelType)))
{
this.type = t;
this.Solve();
this.CalcError();
double rt = this.Size() * Error(Global.confidence);
if (min > rt) { min = rt; best = t; }
}
type = best;
this.Solve();
this.CalcError();
ComputeErrorRange();
error = Error(Global.confidence);
this.len = ts.Length;
}