File size: 5,419 Bytes
e4b9a7b | 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 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 | # Word Tokenization Benchmark for Thai (obsolete)
A framework for benchmarking tokenization algorithms for Thai.
It has a command-line interface that allows users to conveniently execute the benchmarks
as well as a module interface for later use in their development pipelines.
## Metrics
<div align="center">
<img src="https://i.imgur.com/jVBOLa2.png"/>
</div>
### Character-Level (CL)
- True Positive (TP): no. of starting characters that are correctly predicted.
- True Negative (TN): no. of non-starting characters that are correctly predicted.
- False Positive (FP): no. of non-starting characters that are wrongly predicted as starting characters.
- False Negative (FN): no. of starting characters that are wrongly predicted as non-starting characters.
- Precision: TP / (TP + FP)
- Recall: TP / (TP+FN)
- f1: ...
### Word-Level (WL)
- Correctly Tokenized Words (CTW): no. of words in reference that are correctly tokenized.
- Precision: CTW / no. words in reference solution
- Recall: CTW / no. words in sample
- f1: ...
## Benchmark Results
| Vendor | Approach | Datasets |
|---|---|---|
| DeepCut | CNN | [-yellow.svg)][res-BEST-val-DeepCut] [-yellow.svg)][res-THNC-DeepCut] [-yellow.svg)][res-Orchid-DeepCut] [-yellow.svg)][res-WiseSight160-DeepCut] |
| PyThaiNLP-newmm | dictionary-based | [-yellow.svg)][res-BEST-val-PyThaiNLP-newmm] [-yellow.svg)][res-THNC-PyThaiNLP-newmm] [-yellow.svg)][res-Orchid-PyThaiNLP-newmm] [-yellow.svg)][res-WiseSight160-PyThaiNLP-newmm] |
| Sertis-BiGRU | Bi-directional RNN | [-yellow.svg)][res-BEST-val-Sertis-BiGRU] [-yellow.svg)][res-WiseSight160-Sertis-BiGRU] |
[res-BEST-val-DeepCut]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=BEST-val-DeepCut
[res-THNC-DeepCut]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=THNC-DeepCut
[res-Orchid-DeepCut]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=Orchid-DeepCut
[res-WiseSight160-DeepCut]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=WiseSight160-DeepCut
[res-BEST-val-PyThaiNLP-newmm]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=BEST-val-PyThaiNLP-newmm
[res-THNC-PyThaiNLP-newmm]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=THNC-PyThaiNLP-newmm
[res-Orchid-PyThaiNLP-newmm]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=Orchid-PyThaiNLP-newmm
[res-WiseSight160-PyThaiNLP-newmm]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=WiseSight160-PyThaiNLP-newmm
[res-BEST-val-Sertis-BiGRU]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=BEST-val-Sertis-BiGRU
[res-WiseSight160-Sertis-BiGRU]: https://pythainlp.org/tokenization-benchmark-visualization/?experiment-name=WiseSight160-Sertis-BiGRU
## Installation (WIP)
```shell
pip ...
```
## Usages (to be updated)
1. Command-line Interface
```shell
PYTHONPATH=`pwd` python scripts/thai-tokenisation-benchmark.py \
--test-file ./data/best-2010/TEST_100K_ANS.txt \
--input ./data/best-2010-syllable.txt
```
Sample output:
```text
Benchmarking ./data/best-2010-deepcut.txt against ./data/best-2010/TEST_100K_ANS.txt with 2252 samples in total
============== Benchmark Result ==============
metric mean±std min max
char_level:tp 47.82±47.22 1.000000 354.0
char_level:tn 144.19±145.97 1.000000 887.0
char_level:fp 1.34±2.02 0.000000 23.0
char_level:fn 0.70±1.19 0.000000 14.0
char_level:precision 0.96±0.08 0.250000 1.0
char_level:recall 0.98±0.04 0.500000 1.0
char_level:f1 0.97±0.06 0.333333 1.0
word_level:precision 0.92±0.14 0.000000 1.0
word_level:recall 0.93±0.12 0.000000 1.0
word_level:f1 0.93±0.13 0.000000 1.0
```
2. Module Interface
```python
from pythainlp.benchmarks import word_tokenisation as bwt
ref_samples = array of reference tokenised samples
tokenised_samples = array of tokenised samples, aka. from your algorithm
# dataframe contains metrics for each sample
df = bwt.benchmark(ref_samples, tokenised_samples)
```
## Related Work
- [Thai Tokenizers Docker][docker]: collection of Docker containers of pre-built Thai tokenizers.
## Development
Unit tests
```shell
TEST_VERBOSE=1 PYTHONPATH=. python tests/__init__.py
```
## Acknowledgement
This project was initially started by [Pattarawat Chormai][pat], while he was interning at [Dr. Attapol Thamrongrattanarit][ate]'s lab.
[docker]: https://github.com/PyThaiNLP/docker-thai-tokenizers
[ate]: https://attapol.github.io
[pat]: https://pat.chormai.org
|