Context Tree Weighting

Investigating and exploring context-tree weighting

Explaining CTW is a little tough
project presentation

btw here's the code for CTW !

Huffman Coding just to create a baseline

An example of how huffman coding works

Characters	ASCII	Frequency	HUFFMAN	bits
'A'	065	1	00	01000001
'g'	103	1	01	01100111
'i'	108	1	11	01101100
'o'	111	1	10	01101111

This is an image illustrating how huffman coding works (source: WIKI)

And here's the code for Huffman Coding compressor and decompressor

Results for the compression are

algorithms	size(bytes)
ORIGINAL (ENWIK 4)	100326
GZIP (LZ77 and LZ78)	26780
HUFFMAN	53967
7ZIP (LZMA and LZMA2)	23806

AutoEncoder (for image compression)

Assume that there is an explaination for autoencoder here

results for compression with AutoEncoder

assume that there is a table of results here

here is the code for AutoEncoder and the PreTrained Model

Applications

---

Using compression tool to study mammalian evolution and construct the phylogeny of the SARS virus can be found here.

BARF this thing compresses your data to less than 1 byte (i don't know how)

---

The Barf Thingy can be found here.

References

---

• An Autoencoder-based Learned Image Compressor: Description of Challenge Proposal by NCTU
• Text Encryption with Huffman Compression
• The context-tree weighting method: basic properties
• Context Tree Switching

TODO

• Hutter prize
• improve the compression
• manupulate the huffman thing for the project !
• Audio compression (converting it to spectograms can help maybe)

Huffman TODO

• read the compression part
• decompress does not work maybe
• make it visual with some python lib