public NgramBufferBuilder(int ngramLength, int skipLength, int slotLim, NgramIdFinder finder)
{
Contracts.Assert(ngramLength > 0);
Contracts.Assert(skipLength >= 0);
Contracts.Assert(ngramLength <= MaxSkipNgramLength - skipLength);
Contracts.Assert(slotLim >= 0);
_ngramLength = ngramLength;
_skipLength = skipLength;
_slotLim = slotLim;
_ngram = new uint[_ngramLength];
_queue = new FixedSizeQueue <uint>(_ngramLength + _skipLength);
_bldr = BufferBuilder <Float> .CreateDefault();
_finder = finder;
}
public NgramIdFinder Decorate(int iinfo, NgramIdFinder finder)
{
Contracts.Assert(0 <= iinfo && iinfo < _parent._bindings.InfoCount);
Contracts.Assert(_iinfoToCollector[iinfo] == null);
Contracts.AssertValue(finder);
var srcIndices = _parent._bindings.Infos[iinfo].SrcIndices;
// Define the mapper from the ngram, to text.
ValueMapper <NGram, StringBuilder> stringMapper;
StringBuilder temp = null;
char[] buffer = null;
if (srcIndices.Length == 1)
{
// No need to include the column name. This will just be "A" or "(A,B,C)" depending
// on the n-arity of the ngram.
var srcMap = _srcTextGetters[srcIndices[0]];
Contracts.AssertValue(srcMap);
stringMapper =
(ref NGram src, ref StringBuilder dst) =>
{
Contracts.Assert(src.ISrcCol == 0);
if (src.Lim == 1)
{
srcMap(ref src.Grams[0], ref dst);
return;
}
ClearDst(ref dst);
for (int i = 0; i < src.Lim; ++i)
{
if (i > 0)
{
dst.Append('|');
}
srcMap(ref src.Grams[i], ref temp);
InvertHashUtils.AppendToEnd(temp, dst, ref buffer);
}
};
}
else
{
Contracts.Assert(srcIndices.Length > 1);
string[] srcNames = _friendlyNames[iinfo];
if (srcNames == null)
{
srcNames = new string[srcIndices.Length];
for (int i = 0; i < srcIndices.Length; ++i)
{
srcNames[i] = _parent.Source.Schema.GetColumnName(srcIndices[i]);
}
}
Contracts.Assert(Utils.Size(srcNames) == srcIndices.Length);
string[] friendlyNames = _friendlyNames?[iinfo];
// We need to disambiguate the column name. This will be the same as the above format,
// just instead of "<Stuff>" it would be with "ColumnName:<Stuff>".
stringMapper =
(ref NGram src, ref StringBuilder dst) =>
{
var srcMap = _srcTextGetters[srcIndices[src.ISrcCol]];
Contracts.AssertValue(srcMap);
ClearDst(ref dst);
dst.Append(srcNames[src.ISrcCol]);
dst.Append(':');
for (int i = 0; i < src.Lim; ++i)
{
if (i > 0)
{
dst.Append('|');
}
srcMap(ref src.Grams[i], ref temp);
InvertHashUtils.AppendToEnd(temp, dst, ref buffer);
}
};
}
var collector = _iinfoToCollector[iinfo] = new InvertHashCollector <NGram>(
_parent._bindings.Types[iinfo].VectorSize, _invertHashMaxCounts[iinfo],
stringMapper, EqualityComparer <NGram> .Default, (ref NGram src, ref NGram dst) => dst = src.Clone());
return
((uint[] ngram, int lim, int icol, ref bool more) =>
{
Contracts.Assert(0 <= icol && icol < srcIndices.Length);
Contracts.AssertValue(_srcTextGetters[srcIndices[icol]]);
var result = finder(ngram, lim, icol, ref more);
// For the hashing NgramIdFinder, a result of -1 indicates that
// a slot does not exist for the given ngram. We do not pass ngrams
// that do not have a slot to the InvertHash collector.
if (result != -1)
{
// The following ngram is "unsafe", in that the ngram array is actually
// re-used. The collector will utilize its copier to make it safe, in
// the event that this is a key it needs to keep.
var ngramObj = new NGram(ngram, lim, icol);
collector.Add(result, ngramObj);
}
return result;
});
}
