       
 


   ,        .

 ,            ,  AI/ML        .     Jupyter Notebook    ,   ,  .

        .     ,     "  ",    .        ML-:   -       Kubernetes     (Data Drift).

    :    ,     .      ,    MLOps   





 

       





    ,        Jupyter Notebook,   ,      .  ,  ,  AI/ML Engineer     ,   PyTorch,  ,   , ,      .



   ,       ,    ,     .  ,     ,       ,      .  ,     GPU,    ,    .



           .        ML-:   -      .



 I: .     

 1.  :     

1.1.   AI/ML Engineer 

 2012 ,  AlexNet  ImageNet,      "ML Engineer".  ,   ,  ,   .     :     -,       .



 AI/ML Engineer    .  :



:    ,   ,   .



 :   ,  ,   API,    Git.



:     Kubernetes,   CI/CD,   .



:    " "   ROC-AUC,   stakeholder',   .



  Data Scientist  ML Engineer:



      Data Scientist      ML Engineer

      ,        

      Jupyter Notebook      Production-ready 

      Accuracy, F1      Latency, Throughput, Cost

      Pandas, Scikit-learn      PyTorch, Docker, Kubernetes

      ,        

1.2.   ML- (MLLC)

ML-     .  ,     .



 1:  

    : ",     ".       ML: "  .    ROC-AUC > 0.85.    < 100 ".



 2:    

80% .  ,     5  ,  30%  ,      .



 3:   

   ,  feature engineering,  .   Jupyter Notebook.



 4:   

    .  API,  CI/CD,   .



 5:   

   ,  Data Drift,  .



1.3.  :    

 : "    ". ML Engineer : " .   NDCG@10.   cold start   ".



 :



-      ML-      

              ROC-AUC, Recall

             NDCG, MAP

             MAE, RMSE

        detection      Precision@K

             Silhouette Score

 :      -.    Accuracy,      False Negative      .



 2.    

2.1.     

     .       ,  .



  :



python

import numpy as np



# -

v = np.array([[1], [2], [3]])  # shape (3, 1)



# 

W = np.array([[1, 2], [3, 4], [5, 6]])  # shape (3, 2)



#    

result = W.T @ v  # shape (2, 1)

:

          .



: 0D 



: 1D 



: 2D 



 : 4D  (batch, channels, height, width)



,   :



 :   .



 :   .



:   .



Broadcasting:   .



2.2.    

  ?

       .      ,    .



 :



: ,   .



 :  .



:  .



:   .



 :

P( | ) = P( | ) * P() / P()



    ,     .



2.3.  :    PyTorch

:

     L(?).    :



?_{t+1} = ?_t - ? * ?L(?_t)



 ?  learning rate.



 PyTorch:



python

import torch

import torch.nn as nn



model = nn.Linear(10, 1)

optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

criterion = nn.MSELoss()



# Forward pass

y_pred = model(X)

loss = criterion(y_pred, y_true)



# Backward pass

optimizer.zero_grad()

loss.backward()

optimizer.step()

  :



Batch GD:   . ,  .




  .


   .

   ,     (https://www.litres.ru/book/turist-martovskiy/arkhitektory-sinteza-inzheneriia-intellekta-ot-idei-do-p-74609738/)  .

      Visa, MasterCard, Maestro,    ,   ,     ,  PayPal, WebMoney, ., QIWI ,       .


