for proper estimation of symbol's weights when using dictionary compression. Note : using only huffman costs is not good enough, presumably because sequence symbol costs are incorrect.