algorithms
syntax
time complexity
famous names
output
100

Repeatedly compares and swaps adjacent element

bubble sort

100

python: outputing hello world

console.log("Hello, World!")

print("Hello, World!")

100

arr = [i for i in range(int(input()))]

for item in arr:

    print(item)

O(n)

100

guy who made computer or something, something about windows

bill gates

100

x = 5

y = 3

print(x + y)


8

200

Divides the array into halves, sorts them, and combines them

merge sort

200

def greet(name):

    if name == "Alice"

        print("Hello, Alice!")

    else

        print("Hello, stranger!")

where are the colons: 💀💀💀💀💀💀💀💀

200

def find_pairs(arr, target):

    pairs = []

    for i in range(len(arr)):

        for j in range(i + 1, len(arr)):

            if arr[i] + arr[j] == target:

                pairs.append((arr[i], arr[j]))

    return pairs


o(n^2)

200

amazon guy

jeff bezos

200

def mystery(a, b=2, c=3):

    return a + b * c


result = mystery(4, c=5)

print(result)


14

300

Uses a pivot to partition the array into smaller subarrays

quick sort

300

def divide_numbers(a, b):

    try:

        result = a / b

    except valueError as e:

        print("An error occurred:", e)

    return result

ValueError

300

def maga_sort(arr): #CHATGPT

    if len(arr) > 1:

        # Find the middle point and divide the array

        mid = len(arr) // 2

        left_half = arr[:mid]

        right_half = arr[mid:]

        # Recursively sort both halves

        merge_sort(left_half)

        merge_sort(right_half)

        # Merge the sorted halves

        i = j = k = 0

        # Copy data to temp arrays left_half and right_half

        while i < len(left_half) and j < len(right_half):

            if left_half[i] < right_half[j]:

                arr[k] = left_half[i]

                i += 1

            else:

                arr[k] = right_half[j]

                j += 1

            k += 1

        # Check if any element was left

        while i < len(left_half):

            arr[k] = left_half[i]

            i += 1

            k += 1

        while j < len(right_half):

            arr[k] = right_half[j]

            j += 1

            k += 1

o(nlog(n))

300

guy who made the shortest path problem

Dijkstra

300

def func(x, y=[]):

    y.append(x)

    return y


a = func(1)

b = func(2, [])

c = func(3, b)


print(a)

print(b)

print(c)


[1]
[2]
[3, 2]

400

Builds a minimum spanning tree starting from a node

Prim's algorithm

400

arr = [int(input()) for i in range(int(input()))]

arr = sorted(arr, lambda x: x[0])

print(arr)

for item in arr:

    item += 1

print(arr)

for i in range(len(list(map(str, arr)))):

    arr[i] = arr[i] >> 1

print(arr)

you can't take an index of a number 

400

import math

def matrix_chain_multiplication_with_exponentiation(p):

    n = len(p) - 1  # Number of matrices

    dp = [[0] * n for _ in range(n)]

    for chain_len in range(2, n + 1):  # chain_len is the length of the chain

        for i in range(n - chain_len + 1):

            j = i + chain_len - 1

            dp[i][j] = math.inf  # Start with a large number (infinity)

            for k in range(i, j):

                q = dp[i][k] + dp[k + 1][j] + p[i] * p[k + 1] * p[j + 1]

                q += p[k + 1] * p[k + 1]  # Exponentiation cost for matrix A_k

                if q < dp[i][j]:

                    dp[i][j] = q

    return dp[0][n - 1]

O(n^2) or O(n^3)

400

guy who solved the shortest path problem but there are negative numbers for some reason

Richard Bellman and Lester Ford Jr the robbers of Alfonso Shimbel  

400

def outer(x):

    def inner(y):

        nonlocal x

        x += y

        return x

    return inner

a = outer(5)

b = outer(10)

result1 = a(3)

result2 = a(4)

result3 = b(2)

result4 = a(1)

print(result1)

print(result2)

print(result3)

print(result4)


8

12

12

13


500

Combines heuristics and pathfinding for optimal shortest path

A*

500

def process_data(data):

    result = [] # return results

    for i in range(len(data)): # loop for each element

        if isinstance(data[i], int): # if number

            result.append(data[i] * 2) # double

        elif isinstance(data[i], str): # if str

            result.append(data[i].upper()) #capitalize

        else: # else add

            result.append(data[i])

    return result # give back results

doesn't consider edge case float

500

from itertools import combinations

def find_tetrahedral_routes(n, m, roads, T):

    # Initialize the adjacency matrix with high values to represent no road between cities

    INF = float('inf')

    adj_matrix = [[INF] * n for _ in range(n)]

    

    # Fill the adjacency matrix with the given roads and their costs

    for u, v, c in roads:

        adj_matrix[u-1][v-1] = c

        adj_matrix[v-1][u-1] = c

    

    # To store the count of valid Tetrahedral Routes and their total cost

    valid_routes_count = 0

    total_cost = 0

    

    # Iterate through all combinations of 4 cities

    for cities in combinations(range(n), 4):

        a, b, c, d = cities

        

        # Check if all pairs of cities are connected (forming a complete subgraph)

        road_costs = [

            adj_matrix[a][b], adj_matrix[a][c], adj_matrix[a][d],

            adj_matrix[b][c], adj_matrix[b][d], adj_matrix[c][d]

        ]

        

        # If any of the roads do not exist (cost is INF), skip this combination

        if any(cost == INF for cost in road_costs):

            continue

        

        # Calculate the total cost of the roads in this Tetrahedral Route

        total_route_cost = sum(road_costs)

        

        # If the cost is less than or equal to the threshold, it's a valid route

        if total_route_cost <= T:

            valid_routes_count += 1

            total_cost += total_route_cost

    

    return valid_routes_count, total_cost


# Example usage:

n = 6  # Number of cities

m = 9  # Number of roads

roads = [

    (1, 2, 1), (1, 3, 2), (1, 4, 3), (1, 5, 4), (1, 6, 5),

    (2, 3, 1), (2, 4, 2), (2, 5, 3), (2, 6, 4),

]

T = 10  # Threshold cost for Tetrahedral Routes


valid_routes_count, total_cost = find_tetrahedral_routes(n, m, roads, T)

print(valid_routes_count, total_cost)


O(n^2) or O(n^4)

500

Developed the onboard flight software for NASA's Apollo missions. 

Margaret Hamilton

500

def decorator(func):

    def wrapper(*args, **kwargs):

        result = func(*args, **kwargs)

        return result * 2

    return wrapper


@decorator

def multiply(a, b):

    return a * b


@decorator

def add(a, b):

    return a + b


result1 = multiply(2, 3)

result2 = add(2, 3)

result3 = multiply(result1, result2)

result4 = add(result1, result2)


print(result1)

print(result2)

print(result3)

print(result4)


12

10

240

44


M
e
n
u