SecondYearProject/full_run_script.py at main · borchand/SecondYearProject · GitHub

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
from TokenClassificationTrainer import TokenClassificationTrainer
import pandas as pd
import numpy as np
import torch
import pickle


def runs(discriminate_lr = False, save_name = "baseline", rate = 0.7):
    # Set the task and name of the pretrained model and the batch size for finetuning
    task = "ner"
    model_name = "xlm-mlm-17-1280"  # "bert-base-multilingual-cased" or "xlm-mlm-17-1280"
    seed = np.random.randint(0,100000)
    save_name = save_name + ".seed-" + str(seed)
    batch_size = 32

    # Flag to indicate whether to label all tokens or just the first token of each word
    label_all_tokens = False

    # File paths to splits of the chosen dataset
    file_paths = {
        "train": "data/datasets/NoSta-D/NER-de-train.tsv",
        "validation": "data/datasets/NoSta-D/NER-de-dev.tsv",
    }

    # initialize trainer
    trainer = TokenClassificationTrainer(task, model_name, save_name, batch_size, label_all_tokens, file_paths)


    # Training
    trainer.train(discriminate_lr = discriminate_lr, seed = seed,learning_rate=6e-6,rate=rate)

    evals = trainer.evaluate_multiple(["data/datasets/ewt/en_ewt_nn_test_newsgroup_and_weblogs.conll", "data/datasets/NoSta-D/NER-de-test.tsv", "data/datasets/DaNplus/da_news_comb_test.tsv", "data/datasets/hungarian/hungarian_test.tsv"])

    baseline_eval_baseline_model = evals[0]
    NoStaD_eval_baseline_model = evals[1]
    DaNplus_eval_baseline_model = evals[2]
    Hungarian_eval_baseline_model = evals[3]

    cols = ["Dataset", "Language", "Seed"] + [name for name, _ in baseline_eval_baseline_model.items()]

    df = pd.DataFrame(columns=cols)


    trainer.del_trainer()
    del trainer

    # Add the evals to df
    df.loc[0] = ["Baseline", "English", seed] + [value for _, value in baseline_eval_baseline_model.items()]
    df.loc[1] = ["NoSta-D", "German", seed] + [value for _, value in NoStaD_eval_baseline_model.items()]
    df.loc[2] = ["DaNplus", "Danish", seed] + [value for _, value in DaNplus_eval_baseline_model.items()]
    df.loc[3] = ["Hungarian", "Hungarian", seed] + [value for _, value in Hungarian_eval_baseline_model.items()]


    torch.cuda.empty_cache()
    torch.cuda.reset_max_memory_allocated()
    torch.cuda.reset_peak_memory_stats()
    return df


if __name__ == "__main__":
    list_baseline = []
    list_discriminate_lr = []
    list_german_val = []
    list_eng_german_dataset = []
    with open('its working', 'wb') as f:
        pickle.dump('hahaha',f)
    for i in range(5):

        list_baseline.append(runs(save_name = "baseline"))
        with open('list_baseline', 'wb') as f:
            pickle.dump(list_baseline,f)

        list_discriminate_lr.append(runs(discriminate_lr = True, save_name = "discriminate-lr"))
        with open('list_discriminate_lr_new', 'wb') as f:
            pickle.dump(list_discriminate_lr,f)