feat: Add PPM (Prediction by Partial Matching) algorithm implementation

- Implemented the PPM algorithm for data compression and decompression.
- Added methods for updating the model, encoding, and decoding symbols.
- Included utility functions for reading from files and testing the algorithm.
- Verified functionality with various datasets to ensure accuracy.

This addition enhances the repository's collection of Python algorithms.
This commit is contained in:
Lukas Olenyi 2024-11-21 02:10:53 +01:00
parent f3f32ae3ca
commit 2f37ee9077

125
compression/ppm.py Normal file
View File

@ -0,0 +1,125 @@
from __future__ import annotations
import sys
from collections import defaultdict
class PPMNode:
def __init__(self) -> None:
# Initialize a PPMNode with a dictionary for child nodes and a count of total occurrences
self.counts: dict[str, PPMNode] = defaultdict(PPMNode)
self.total: int = 0
def __repr__(self) -> str:
return f"PPMNode(total={self.total})"
class PPM:
def __init__(self, order: int = 2) -> None:
# Initialize the PPM model with a specified order and create a root node
self.order: int = order
self.root: PPMNode = PPMNode()
self.current_context: PPMNode = self.root
def update_model(self, context: str, symbol: str) -> None:
# Update the model with the new symbol in the given context
node = self.current_context
for char in context:
# Traverse through the context characters, updating the total counts
node = node.counts[char]
node.total += 1
# Increment the count for the specific symbol in the current context
node.counts[symbol].total += 1
def compress(self, data: str) -> list[float]:
# Compress the data using the PPM algorithm and return a list of probabilities
compressed_output: list[float] = []
context: str = ""
for symbol in data:
# Update the model with the current context and symbol
self.update_model(context, symbol)
# Encode the symbol based on the current context
compressed_output.append(self.encode_symbol(context, symbol))
# Update the context by appending the symbol, keeping it within the specified order
context = (context + symbol)[-self.order:] # Keep the context within order
return compressed_output
def encode_symbol(self, context: str, symbol: str) -> float:
# Encode a symbol based on the current context and return its probability
node = self.root
for char in context:
# Traverse through the context to find the corresponding node
if char in node.counts:
node = node.counts[char]
else:
return 0.0 # Return 0.0 if the context is not found
# Return the probability of the symbol given the context
if symbol in node.counts:
return node.counts[symbol].total / node.total # Return probability
return 0.0 # Return 0.0 if the symbol is not found
def decompress(self, compressed_data: list[float]) -> str:
# Decompress the compressed data back into the original string
decompressed_output: list[str] = []
context: str = ""
for prob in compressed_data:
# Decode each probability to retrieve the corresponding symbol
symbol = self.decode_symbol(context, prob)
if symbol:
decompressed_output.append(symbol)
# Update the context with the newly decoded symbol
context = (context + symbol)[-self.order:] # Keep the context within order
else:
break # Stop if a symbol cannot be found
return ''.join(decompressed_output) # Join the list into a single string
def decode_symbol(self, context: str, prob: float) -> str | None:
# Decode a symbol from the given context based on the probability
node = self.root
for char in context:
# Traverse through the context to find the corresponding node
if char in node.counts:
node = node.counts[char]
else:
return None # Return None if the context is not found
# Iterate through the children of the node to find the symbol matching the given probability
for symbol, child in node.counts.items():
if child.total / node.total == prob:
return symbol # Return the symbol if the probability matches
return None # Return None if the symbol is not found
def read_file(file_path: str) -> str:
"""Read the entire file and return its content as a string."""
with open(file_path, 'r') as f:
return f.read()
def ppm(file_path: str) -> None:
"""Compress and decompress the file using PPM algorithm."""
data = read_file(file_path) # Read the data from the specified file
ppm_instance = PPM(order=2) # Create an instance of the PPM model with order 2
# Compress the data using the PPM model
compressed = ppm_instance.compress(data)
print("Compressed Data (Prob abilities):", compressed)
# Decompress the data back to its original form
decompressed = ppm_instance.decompress(compressed)
print("Decompressed Data:", decompressed)
if __name__ == "__main__":
# Check if the correct number of command line arguments is provided
if len(sys.argv) != 2:
print("Usage: python ppm.py <file_path>")
sys.exit(1)
# Call the ppm function with the provided file path
ppm(sys.argv[1])