1) CTO introduces the company 

2) Tell about yourself 

3) Experience with organizing annotation workflow and managing responsibilities

1) Can you use accuracy as a loss function?
2) Can you use IoU as a loss function?
3) What is the difference between binary, multiclass, and multilabel classification?
4) What is the hash map and how is it used in Python?
5) What is the time complexity of insert into a hash map?
6) What is the difference between linked list and array?
7) What is the difference between local and global minima?
8) What is the aim of backpropagation algorithm?
9) How to copy data from a remote location?
10) How to make a program (i.e. training loop) run after session is terminated?

def f(arr: List):
    arr.append(1)
    return arr
arr1 = []
arr2 = arr1
arr3 = f(arr1)
arr4 = f(arr2)
print(arr4)

What will be the output?

def duplicate(arr: List):
    return ...

Create a function to test if the list contains duplicates

def pair_to_sum(nums: List, target:int):
    return ...

Create a function, that will return indices of 2 elements in an array, that sum up to a target, given solution is unique and always exists

1) Suppose a business customer comes to you with a request to build a system that will be able to distinguish between 5 types of lung diseases, but they have only unlabeled data, what will you do?

- How would you label the data?
- Suppose the system is working, but there are some edge cases that are classified incorrectly, how to solve this problem?
- Suppose there are sometime incorrect inputs, i.e. leg x-ray in a lung disease detector, how to handle such cases?
- Suppose a new hospital is about to be connected to the system, how to handle such problem?

- Types of learning (Supervised, Unsupervised, Semi-Supervised)
- Math (Matrices, Probability, Statistics, Distributions)
- Basic ML algorithms (Regression, KNN, Boosting, Decision Trees)
- Scikit-Learn
- Parameter tuning
- Python data processing (numpy, pandas, matplotlib)
- Neural networks (Convolution, Augmentations, Overfitting, Transfer learning, Embeddings, PyTorch, TensorFlow, paperswithcode.com)

import pandas as pd


df.assign(column_one = lambda row: row["existing_column"] // 1024,
column_two = 24,
....)

import numpy as np
import pandas as pd


def calculate(df: pd.DataFrame, thr: float = 2.56, columns: list = []):
    for col in columns:
       df[f"{col}_transformed"] = np.sqrt(df[col]) > 2.56
    return df

df.pipe(calculate, thr=5, columns=["value_1", "value_2"])

from collections import defaultdict

res = defaultdict(list) # общая структура словарь, где для каждого ключа значением по-умолчанию будет пустой список

for idx, value enumerate([123, 999, 678]):
    res["idx"].append(idx)
    res["value"].append(value**2)

res["value"] # это список =)

import typing
import pandas as pd
import numpy as np


def func(df: pd.DataFrame, columns: list, method: t.Callable = np.mean, window_width: int = 2):
    assert window_width == 2, "Ширина окна должна быть 2+!"

    for col in columns:
        df[f"{col}_rolling"] = df[col].rolling(window_width, min_periods=2).apply(lambda window: method(window))

    return df

func_mean = partial(func, columns=["value_1", "value_2"], method=np.median)

# теперь можно так, остальные параметры уже частично применились выше
func_mean(df=df)

l = [0, 1, -5, 7, 8]

# x - первый элемент списка, или результат применения функции к текущему элементу, y - текущий элемент
res = reduce(lambda x, y: x+y, l)

print(res) # 11