dayyass

import numpy as np

import torch
import torch.nn as nn

import onnx
import onnxruntime


##### INIT 1d, 2d, 3d GLOBAL POOLING MODULES #####

dayyass

"""
In a binary classification problem, a neural network usually returns a vector of logits of shape [batch_size],
while in a multiclass classification problem, logits are represented as a matrix of shape [batch_size, n_classes].

For these tasks, different loss functions are used, and, therefore, the network training pipelines are also different,
which is not convenient when you need to test hypotheses for both problem statements (binary/multiclass).

Pipeline schemes:
- binary classification:
logits (of shape [batch_size]) -> BCEWithLogitsLoss

dayyass

import torch
import random
import numpy as np


def set_global_seed(seed: int):
    """
    Set global seed for reproducibility.
    """

dayyass

dayyass

dayyass

dayyass

import re


# Method build_tokenizer from _VectorizerMixin mixin from which classes HashingVectorizer, CountVectorizer and 
# TfidfVectorizer (through CountVectorizer) are partially inherited.
# It is used to split a string into a sequence of tokens (only if analyzer == 'word').
def build_tokenizer(token_pattern: str = r"(?u)\b\w\w+\b"):
    """
    Return a function that splits a string into a sequence of tokens.
    

dayyass

import numpy as np
from tqdm import trange
from collections import defaultdict
from typing import Dict, Tuple, DefaultDict


def get_matrix_idx_to_value_dict(
    matrix: np.ndarray,
    verbose: bool = True,
) -> DefaultDict[Tuple[int, int], int]:

dayyass

- repo: local
  hooks:
    - id: unittest
      name: unittest
      entry: python -m unittest discover
      language: python
      always_run: true
      pass_filenames: false

dayyass

FROM cr.msk.sbercloud.ru/aicloud-jupyter/jupyter-cuda10.1-tf2.2.0-mlspace:latest
MAINTAINER Dani El-Ayyass <[email protected]>

USER root

# Docker
# Set up the repository
RUN apt-get update
RUN apt-get -y install apt-transport-https ca-certificates curl gnupg lsb-release