mirror of
https://github.com/onyx-and-iris/grokking-algorithms.git
synced 2024-11-16 01:40:53 +00:00
67 lines
1.7 KiB
Python
67 lines
1.7 KiB
Python
|
import logging
|
||
|
from dataclasses import dataclass
|
||
|
|
||
|
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__)
|
||
|
|
||
|
|
||
|
@dataclass
|
||
|
class Location:
|
||
|
name: str
|
||
|
time: int
|
||
|
rating: int
|
||
|
|
||
|
|
||
|
def dynamic(items, n):
|
||
|
# create table and zero fill it (required for calculations)
|
||
|
table = [[0 for _ in range(2 * (W + 1) - 1)] for _ in range(n + 1)]
|
||
|
|
||
|
# calculate all possible max values for items in knapsack
|
||
|
for i in range(1, n + 1):
|
||
|
for w in range(1, 2 * (W + 1) - 1):
|
||
|
if items[i - 1].time <= w:
|
||
|
table[i][w] = max(
|
||
|
items[i - 1].rating + table[i - 1][w - items[i - 1].time],
|
||
|
table[i - 1][w],
|
||
|
)
|
||
|
else:
|
||
|
table[i][w] = table[i - 1][w]
|
||
|
|
||
|
format_and_print(table)
|
||
|
|
||
|
return table[-1][-1]
|
||
|
|
||
|
|
||
|
def format_and_print(table):
|
||
|
# enumerate first row (headings) + label each row in column0
|
||
|
for i, row in enumerate(table):
|
||
|
if i == 0:
|
||
|
continue
|
||
|
row[0] = locations[i - 1].name
|
||
|
table[0] = [i / 2 for i in range(2 * (W + 1) - 1)]
|
||
|
table[0][0] = None
|
||
|
|
||
|
# print tabularised 2D array
|
||
|
logger.info([item.name for item in locations])
|
||
|
logger.info(tabulate(table, tablefmt="pretty"))
|
||
|
|
||
|
|
||
|
# values converted to integers
|
||
|
locations = [
|
||
|
Location("Westminster Abbey", 1, 7),
|
||
|
Location("Globe Theater", 1, 6),
|
||
|
Location("National Gallery", 2, 9),
|
||
|
Location("British Museum", 4, 9),
|
||
|
Location("St. Pauls Cathedral", 1, 8),
|
||
|
]
|
||
|
W = 2
|
||
|
greatest_value = dynamic(locations, len(locations))
|
||
|
|
||
|
print(f"Greatest value: {greatest_value}")
|