mirror of
https://github.com/onyx-and-iris/grokking-algorithms.git
synced 2024-11-16 01:40:53 +00:00
58 lines
1.3 KiB
Python
58 lines
1.3 KiB
Python
import logging
|
|
|
|
from tabulate import tabulate
|
|
|
|
logging.basicConfig(
|
|
level=logging.DEBUG,
|
|
format="%(asctime)s %(levelname)s\n\r%(message)s",
|
|
datefmt="%H:%M:%S",
|
|
)
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def longest_common_substring(X, Y, len_x, len_y):
|
|
# create table and zero fill it
|
|
table = [[0 for _ in range(len_x + 1)] for _ in range(len_y + 1)]
|
|
|
|
longest = 0
|
|
|
|
for i in range(1, len_y + 1):
|
|
for j in range(1, len_x + 1):
|
|
if X[j - 1] == Y[i - 1]:
|
|
table[i][j] = table[i - 1][j - 1] + 1
|
|
longest = max(longest, table[i][j])
|
|
else:
|
|
table[i][j] = 0
|
|
|
|
format_and_print(X, Y, table)
|
|
|
|
return longest
|
|
|
|
|
|
def format_and_print(X, Y, table):
|
|
# enumerate first row (headings) + label each row in column0
|
|
for i, row in enumerate(table):
|
|
if i == 0:
|
|
continue
|
|
row[0] = Y[i - 1]
|
|
table[0] = [None, *X]
|
|
|
|
# print tabularised 2D array
|
|
logger.info(tabulate(table, tablefmt="pretty"))
|
|
|
|
|
|
X = "fish"
|
|
Y = "vish"
|
|
|
|
logger.info(f"words: {X} {Y}")
|
|
longest_substring = longest_common_substring(X, Y, len(X), len(Y))
|
|
print(f"Longest common substring: {longest_substring}")
|
|
|
|
|
|
X = "vista"
|
|
Y = "hish"
|
|
|
|
logger.info(f"words: {X} {Y}")
|
|
longest_substring = longest_common_substring(X, Y, len(X), len(Y))
|
|
print(f"Longest common substring: {longest_substring}")
|