text-classification/evalhelper.py at master · silknow/text-classification · 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
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
115
116
117
118
119
120
121
122
123
124
125
126
"""
Model evaluation helper functions + classification functions
"""
import os
import json
import numpy as np
import pandas as pd
from sklearn.metrics import classification_report, accuracy_score
from sklearn.metrics import precision_recall_fscore_support


def hamming_score(y_true, y_pred, normalize=True, sample_weight=None):
    """Compute the Hamming score (a.k.a. label-based accuracy)
    for the multi-label case: http://stackoverflow.com/q/32239577/395857
    """
    acc_list = []
    for i in range(y_true.shape[0]):
        set_true = set(np.where(y_true[i])[0])
        set_pred = set(np.where(y_pred[i])[0])
        tmp_a = None
        if len(set_true) == 0 and len(set_pred) == 0:
            tmp_a = 1
        else:
            tmp_a = len(set_true.intersection(set_pred)) / float(
                len(set_true.union(set_pred))
            )
        acc_list.append(tmp_a)
    return np.mean(acc_list)


def report(y_true, y_pred, target_names):
    report_pred = ""
    labels = list(set(list(y_true)))

    report_pred = classification_report(
        y_true, y_pred, labels=labels, target_names=target_names
    )

    acc_pred = accuracy_score(y_true, y_pred)
    report_pred = report_pred + "\naccuracy:\t{}".format(acc_pred)

    return report_pred


def simple_accuracy(preds, labels):
    return (preds == labels).mean()


def save_errors(dst, y_true, y_pred, target_names, dataset):
    y_true = list(y_true)
    y_pred = list(y_pred)

    rows = []
    for ii, (t, p) in enumerate(zip(y_true, y_pred)):
        if t == p:
            continue
        t = target_names[t]
        p = target_names[p]
        text, label = dataset.reverse(ii)
        assert label == t
        source = dataset.df.iloc[ii]["dataset"]
        filename = dataset.df.iloc[ii]["filename"]
        rows.append(
            {
                "filename": filename,
                "dataset": source,
                "input": text,
                "pred": p,
                "true": t,
            }
        )
    df = pd.DataFrame(rows)
    df.to_csv(dst, sep="\t")


def save_classification(
    dst, y_pred, target_names, dataset, scores=None, ids=None
):
    y_pred = list(y_pred)

    rows = []
    for ii, p in enumerate(y_pred):
        p = target_names[p]
        text, _ = dataset.reverse(ii)
        row = {"txt": text, "predicted": p}

        if scores is not None:
            row["score"] = scores[ii]
        if ids is not None:
            row["id"] = ids[ii]
        rows.append(row)
    df = pd.DataFrame(rows)
    print(f"Saving to {dst} ({len(df)})")
    df.to_csv(dst, sep="\t", index=False)


def results_append(dst, y_true, y_pred, labels, name, task):
    acc = accuracy_score(y_true, y_pred)

    r = report(y_true, y_pred, labels)
    lbl_true = [labels[v] for v in list(y_true)]
    lbl_pred = [labels[v] for v in list(y_pred)]
    pr, rc, f1, sp = precision_recall_fscore_support(
        lbl_true, lbl_pred, beta=1.0, labels=labels, average="micro"
    )
    res = {}
    if os.path.isfile(dst):
        res = json.load(open(dst))

    if not res.get(name):
        res[name] = {}
    res[name][task] = {
        "accuracy": acc,
        "f1": f1,
        "precision": pr,
        "recall": rc,
        "report": r,
    }
    with open(dst, "w") as outfile:
        json.dump(res, outfile)


def save_hidden_vectors(dst, hidden):
    with open(dst, "w") as fout:
        for x in hidden:
            np.savetxt(fout, x)