def encode_block(in_bytes):
if len(in_bytes) == 1:
block = BitArray()
block.append(Bits('0b1'))
block.append(in_bytes)
return block.tobytes()
bw_xf, eof_idx = burrows_wheeler_transform(in_bytes)
front_xf = move_to_front_transform(bw_xf)
rle_data = run_length_encode(front_xf)
huff_data, huff_symbols, serialized_tree = huffman_encode(rle_data,
symbol_bits=8)
huff_len = len(huff_data)
tree_len = len(serialized_tree)
block = BitArray()
block.append(Bits('0b0'))
block.append(Bits(uint=tree_len, length=16))
block.append(Bits(serialized_tree))
block.append(Bits(uint=huff_symbols, length=16))
block.append(Bits(uint=huff_len, length=16))
block.append(Bits(huff_data))
block.append(Bits(uint=eof_idx, length=BLOCK_SIZE_BITS))
return block.tobytes()
def lz77huff_encode(input_data, window_bits=DEFAULT_WINDOW_BITS):
tokens = lz77_encode_to_tokens(input_data, window_bits)
symbols = [s for tok in tokens for s in tok]
bits = BitArray()
for s in symbols:
bits.append(Bits(uint=s, length=window_bits))
encoded_data, num_symbols, serialized_tree = huffman_encode(bits.tobytes(),
symbol_bits=window_bits)
return encoded_data, num_symbols, serialized_tree
def main():
if len(sys.argv) < 3:
print("\nUsage: python3 driver.py <inFilename> <outFilename>\n"
"Include \'.txt\' extension on filenames.\n")
quit()
in_file = sys.argv[1]
out_file = sys.argv[2]
huffman.huffman_encode(in_file, out_file)
input_bit_count = huffman.char_count(in_file)*8
output_bit_count = huffman.char_count(out_file)
print('\nFile encoded. The original file contained ' + str(input_bit_count) + ' bits, while '
'the encoded file contains ' + str(output_bit_count) + ' bits. The original file was '
'reduced to ' + str(round(float(output_bit_count/input_bit_count)*100, 1)) + ' % of '
'it\'s original size.\n')
for s, bits in codes.items():
print "\t%s \t%s" % (s, bits)
# Entropy and mean bits per symbol:
ent = entropy(freq)
mbs = mean_bits(freq, codes)
print "entropy ", ent
print "mean bits", mbs
# Compression ratio:
cr = CHARSIZE / mbs
print "Huffman compression ratio %.2f:1" % cr
print "------------------------"
# Encode the text using the generated huffman codes
encoded = huff.huffman_encode(text, codes)
decoded = huff.huffman_decode(encoded, codes)
# Calcualte number of bits used by the plain text and the huffman encoded bits
bits_text = len(text) * CHARSIZE * 1.0
bits_encoded = len(encoded)
print "text: ", text[:16], "..."
print "encoded: ", encoded[:16], "..."
print "text == decoded:", text == decoded
print "size after compression: %.5f%%" % (bits_encoded / bits_text)
print "NOTE: compression ratio afected by symbol '10' being 2 characters."
def forward(self, input):
if self.transform:
self.channels = self.out_channels
if not hasattr(self, 'mn'):
self.register_buffer('mn', torch.zeros(self.channels, 1))
# Calculate projection matrix if needed
if self.collectStats:
im = F.conv2d(input, self.weight, self.bias, self.stride,
self.padding, self.dilation, self.groups)
N, C, H, W = im.shape # N x C x H x W
im = featuresReshape(im, N, C, H, W, self.microBlockSz,
self.channelsDiv)
self.mn = torch.mean(im, dim=1, keepdim=True)
# Centering the data
im = im - self.mn
self.get_stats_params(im)
# projection
imProj = torch.matmul(self.u.t(), im)
# conv + bn if exists + projection
im2 = F.conv2d(input, self.weight, self.bias, self.stride,
self.padding, self.dilation, self.groups)
imProj2 = featuresReshape(im2, N, C, H, W, self.microBlockSz,
self.channelsDiv)
assert (torch.max(torch.abs(imProj - imProj2)) < 0.1)
else:
# conv + bn if exists + projection
im = F.conv2d(input, self.weight, self.bias, self.stride,
self.padding, self.dilation, self.groups)
N, C, H, W = im.shape # N x C x H x W
imProj = featuresReshape(im, N, C, H, W, self.microBlockSz,
self.channelsDiv)
mult = torch.zeros(1).to(imProj)
add = torch.zeros(1).to(imProj)
self.collectStats = False
dynMax = torch.max(imProj)
dynMin = torch.min(imProj)
if self.actBitwidth < 30:
imProj, mult, add = part_quant(imProj,
max=dynMax,
min=dynMin,
bitwidth=self.actBitwidth)
self.act_size = imProj.numel()
self.bit_per_entry = shannon_entropy(imProj).item()
self.bit_count = self.bit_per_entry * self.act_size
if False: #add if want to show huffman code in additional to theoretical entropy
self.bit_countH = huffman_encode(imProj)
self.bit_per_entryH = self.bit_countH / self.act_size
if self.actBitwidth < 30:
imProj = imProj * mult + add
imProj = torch.matmul(self.u, imProj)
# return original mean
imProj = imProj + self.mn
input = featuresReshapeBack(imProj, N, C, H, W, self.microBlockSz,
self.channelsDiv)
else:
input = F.conv2d(input, self.weight, self.bias, self.stride,
self.padding, self.dilation, self.groups)
return input
def forward(self, input):
if self.transform:
N, C, H, W = input.shape # N x C x H x W
im = featuresReshape(input, N, C, H, W, self.microBlockSz,
self.channelsDiv)
self.channels = im.shape[0]
mn = torch.mean(im, dim=1, keepdim=True)
# Centering the data
im = im - mn
# Calculate projection matrix if needed
if self.collectStats:
self.u, self.s = get_projection_matrix(im, self.transformType,
self.eigenVar)
self.original_channels = self.u.shape[0]
self.channels = self.u.shape[1]
# projection
imProj = torch.matmul(self.u.t(), im)
mult = torch.zeros(1).to(imProj)
add = torch.zeros(1).to(imProj)
dynMax = torch.max(imProj)
dynMin = torch.min(imProj)
if self.actBitwidth < 30:
imProj, mult, add = part_quant(imProj,
max=dynMax,
min=dynMin,
bitwidth=self.actBitwidth)
self.act_size = imProj.numel()
self.bit_per_entry = shannon_entropy(imProj).item()
self.bit_count = self.bit_per_entry * self.act_size
if False: #add if want to show huffman code in additional to theoretical entropy
self.bit_countH = huffman_encode(imProj)
self.bit_per_entryH = self.bit_countH / self.act_size
if self.actBitwidth < 30:
imProj = imProj * mult + add
imProj = torch.matmul(self.u, imProj)
# Bias Correction
imProj = imProj - torch.mean(imProj, dim=1, keepdim=True)
self.mse = torch.sum((imProj - im)**2)
# return original mean
imProj = imProj + mn
# return to general
input = featuresReshapeBack(imProj, N, C, H, W, self.microBlockSz,
self.channelsDiv)
self.collectStats = False
input = self.relu(input)
return input
print 'Start program'
try:
print 'Start open files'
source = open('source.txt')
print 'Open source...OK'
target_lzw = open('lzw_result.txt', 'w')
target_lzw_dict = open('lzw_dict.txt', 'w')
target_huffman = open('huffman_result.txt', 'w')
target_huffman_dict = open('huffman_dict.txt', 'w')
all_text = source.read()
print 'open files...OK'
except IOError as e:
print 'error'
lzw_encode_text, lzw_encode_dict = LZW_encode(all_text)
print 'lzw encode...OK'
target_lzw.write(''.join(lzw_encode_text))
target_lzw_dict.write(''.join(lzw_encode_dict))
print 'lzw file write...OK'
coding = {}
result = huffman_encode(all_text, coding)
print 'huffman encode...OK'
target_huffman.write(''.join(result))
target_huffman_dict.write(''.join(coding))
print 'huffman file write...OK'
target_huffman.close()
target_lzw.close()
target_huffman_dict.close()
target_lzw_dict.close()
def huffman_call(self, widget):
inp_str = self.txt_inp_text.get_text()
out_code, out_encoded = huffman_encode(inp_str)
# print(out_code, out_encoded)
print("Tamanho original da string: ", len(inp_str))
print("Tamanho após codificação: ", len(out_encoded))
def lzw_huff_encode(in_bytes, symbol_len=DEFAULT_SYMBOL_LEN):
lzw_bytes = lzwf_encode(input_data, code_len=symbol_len)
return huffman_encode(lzw_bytes, symbol_bits=symbol_len)
