squad_network

Structure:

• lab-1:

  Linear model and gradient descent

  Basic classification
  Supervised learning basics
  Linear regression
  Validation
  Binary linear classification
  Gradient descent
  Stochastic gradient descent
  

• lab-2:

  Multilayer perceptron

  Linear multi-classification problem
  Universal approximation theorem
  Backpropagation
  Autoencoders
  

• lab-3:

  Convolutional neural network

  Kaggle competition https://www.kaggle.com/c/ch-2017
  

• lab-4:

  Embeddings

    Latent semantic analysis (LSA)
    Word2vec
    GloVe
    FastText
  

• final-lab:

  SQuAD:

    Tree for a working repo
    .
    ├── models
    │   ├── last-v-1.h5
    │   ├── last-v-2.h5
    │   ├── last-v-3.h5
    │   ├── last-v-4.h5
    │   └── last-v-5.h5
    ├── data
    │   ├── context_data.npy
    |   ├── qst_data.npy
    |   ├── cont_features.npy
    │   ├── ans_beg.npy
    |   ├── ans_end.npy
    |   ├── qst_data.npy
    |   ├── con_tag.npy
    |   ├── ent_tag.npy
    │   ├── embedding.npy
    |   ├── context_test.npy
    |   ├── cont_features_test.npy
    |   ├── con_test.npy
    |   ├── ent_test.npy
    |   ├── qst_test.npy
    |   ├── embedding.npy
    |   ├── test_begin.npy
    |   ├── test_end.npy
    |   ├── data.msgpack
    |   └── meta.msgpack
    ├── predict.py
    ├── prepare.py
    ├── test.py
    └── train.py
  

  Firstly, you have to install Spacy lib:

       pip install -U spacy
       python -m spacy download en
  

  Just run this in terminal or visit the website https://spacy.io/usage/

  To train a model run a script named train.py

  To test a model run a script names test.py

  To predict something use predict.py, all instructions are included

  All additional data for directory data (warning it contains large files) you can find on my google drive https://drive.google.com/open?id=1R5ANjuQv26QCczuANHKoYFbvzsp3xGGj

  You have to move data from this archive into final_lab/data

---