File size: 2,439 Bytes
58611ba
 
 
6ff00b6
58611ba
 
6ff00b6
58611ba
 
 
6ff00b6
 
 
 
6e8068e
 
58611ba
 
6ff00b6
58611ba
5bfb9da
58611ba
6ff00b6
58611ba
 
 
 
 
6ff00b6
58611ba
6ff00b6
58611ba
6ff00b6
 
58611ba
6ff00b6
 
58611ba
6ff00b6
 
 
58611ba
6ff00b6
 
58611ba
 
 
 
 
 
6ff00b6
58611ba
 
 
 
 
 
 
 
 
 
 
 
6ff00b6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
---
license: apache-2.0
tags:
- TDA
metrics:
- accuracy
- matthews_correlation
model-index:
- name: bert-base-cased-ru-cola_32_3e-05_lr_0.1_decay_balanced
  results: []
datasets:
- RussianNLP/rucola
language:
- ru
widget:
  - text: Мечтает, и игра очень отличается от последних трёх входов серии.
---

[**Official repository**](https://github.com/upunaprosk/la-tda)

# RuBERT-TDA

This model is a version of [sberbank-ai/ruBert-base](https://huggingface.co/sberbank-ai/ruBert-base) fine-tuned on [RuCoLA](https://huggingface.co/datasets/RussianNLP/rucola).
It achieves the following results on the evaluation set:
- Loss: 1.5267
- Accuracy: 0.8026
- Mcc: 0.4195

## Features extracted from Transformer

The features extracted from attention maps include the following:

1. **Topological features** are properties of attention graphs. Features of directed attention graphs include the number of strongly connected components, edges, simple cycles and average vertex degree. The properties of undirected graphs include
the first two Betti numbers: the number of connected components and the number of simple cycles, the matching number and the chordality.

2. **Features derived from barcodes** include descriptive characteristics of 0/1-dimensional barcodes and reflect the survival (death and birth) of
connected components and edges throughout the filtration.

3. **Distance-to-pattern** features measure the distance between attention matrices and identity matrices of pre-defined attention patterns, such as attention to the first token [CLS] and to the last
[SEP] of the sequence, attention to previous and
next token and to punctuation marks.

The **computed features and barcodes** can be found in the subdirectories of the repository. *test_sub*  features and barcodes were computed on the out-of-domain test RuCoLA dataset.
Refer to notebooks 4* and 5* from the [repository](https://github.com/upunaprosk/la-tda) to construct the classification pipeline with TDA features.

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5.0

### Framework versions

- Transformers 4.27.0.dev0
- Pytorch 1.13.1+cu116
- Datasets 2.9.0
- Tokenizers 0.13.2