anuragshas/en-hi-transliteration

Dataset

NEWS2018 DATASET_04, Task ID: M-EnHi

Notebooks

• xmltodict.ipynb contains the code to convert the xml files to json for training
• training_script.ipynb contains the code for training and inference. It is a modified version of https://github.com/AI4Bharat/IndianNLP-Transliteration/blob/master/NoteBooks/Xlit_TrainingSetup_condensed.ipynb

Predictions

pred_test.json contains top-10 predictions on the validation set of the dataset

Evaluation Scores on validation set

TOP 10 SCORES FOR 1000 SAMPLES

Metrics	Score
ACC	0.703000
Mean F-score	0.949289
MRR	0.486549
MAP_ref	0.381000

TOP 5 SCORES FOR 1000 SAMPLES:

Metrics	Score
ACC	0.621000
Mean F-score	0.937985
MRR	0.475033
MAP_ref	0.381000

TOP 3 SCORES FOR 1000 SAMPLES:

Metrics	Score
ACC	0.560000
Mean F-score	0.927025
MRR	0.461333
MAP_ref	0.381000

TOP 2 SCORES FOR 1000 SAMPLES:

Metrics	Score
ACC	0.502000
Mean F-score	0.913697
MRR	0.442000
MAP_ref	0.381000

TOP 1 SCORES FOR 1000 SAMPLES:

Metrics	Score
ACC	0.382000
Mean F-score	0.881272
MRR	0.382000
MAP_ref	0.380500