File size: 5,248 Bytes
fddf3ff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
127
128
import asyncio
import gc
import logging
import os

import pandas as pd
import psutil
import streamlit as st
from PIL import Image
from streamlit import components
from streamlit.caching import clear_cache
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from transformers_interpret import SequenceClassificationExplainer

os.environ["TOKENIZERS_PARALLELISM"] = "false"
logging.basicConfig(
    format="%(asctime)s : %(levelname)s : %(message)s", level=logging.INFO
)


def print_memory_usage():
    logging.info(f"RAM memory % used: {psutil.virtual_memory()[2]}")


@st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=1)
def load_model(model_name):
    return (
        AutoModelForSequenceClassification.from_pretrained(model_name),
        AutoTokenizer.from_pretrained(model_name),
    )


def main():

    st.title("Transformers Interpet Demo App")

    image = Image.open("./images/tight@1920x_transparent.png")
    st.sidebar.image(image, use_column_width=True)
    st.sidebar.markdown(
        "Check out the package on [Github](https://github.com/cdpierse/transformers-interpret)"
    )
    st.info(
        "Due to limited resources only low memory models are available. Run this [app locally](https://github.com/cdpierse/transformers-interpret-streamlit) to run the full selection of available models. "
    )

    # uncomment the options below to test out the app with a variety of classification models.
    models = {
        # "textattack/distilbert-base-uncased-rotten-tomatoes": "",
        # "textattack/bert-base-uncased-rotten-tomatoes": "",
        # "textattack/roberta-base-rotten-tomatoes": "",
        # "mrm8488/bert-mini-finetuned-age_news-classification": "BERT-Mini finetuned on AG News dataset. Predicts news class (sports/tech/business/world) of text.",
        # "nateraw/bert-base-uncased-ag-news": "BERT finetuned on AG News dataset. Predicts news class (sports/tech/business/world) of text.",
        "distilbert-base-uncased-finetuned-sst-2-english": "DistilBERT model finetuned on SST-2 sentiment analysis task. Predicts positive/negative sentiment.",
        # "ProsusAI/finbert": "BERT model finetuned to predict sentiment of financial text. Finetuned on Financial PhraseBank data. Predicts positive/negative/neutral.",
        "sampathkethineedi/industry-classification": "DistilBERT Model to classify a business description into one of 62 industry tags.",
        "MoritzLaurer/policy-distilbert-7d": "DistilBERT model finetuned to classify text into one of seven political categories.",
        # # "MoritzLaurer/covid-policy-roberta-21": "(Under active development ) RoBERTA model finetuned to identify COVID policy measure classes ",
        # "mrm8488/bert-tiny-finetuned-sms-spam-detection": "Tiny bert model finetuned for spam detection. 0 == not spam, 1 == spam",
    }
    model_name = st.sidebar.selectbox(
        "Choose a classification model", list(models.keys())
    )
    model, tokenizer = load_model(model_name)
    if model_name.startswith("textattack/"):
        model.config.id2label = {0: "NEGATIVE (0) ", 1: "POSITIVE (1)"}
    model.eval()
    cls_explainer = SequenceClassificationExplainer(model=model, tokenizer=tokenizer)
    if cls_explainer.accepts_position_ids:
        emb_type_name = st.sidebar.selectbox(
            "Choose embedding type for attribution.", ["word", "position"]
        )
        if emb_type_name == "word":
            emb_type_num = 0
        if emb_type_name == "position":
            emb_type_num = 1
    else:
        emb_type_num = 0

    explanation_classes = ["predicted"] + list(model.config.label2id.keys())
    explanation_class_choice = st.sidebar.selectbox(
        "Explanation class: The class you would like to explain output with respect to.",
        explanation_classes,
    )
    my_expander = st.beta_expander(
        "Click here for description of models and their tasks"
    )
    with my_expander:
        st.json(models)

    # st.info("Max char limit of 350 (memory management)")
    text = st.text_area(
        "Enter text to be interpreted",
        "I like you, I love you",
        height=400,
        max_chars=850,
    )

    if st.button("Interpret Text"):
        print_memory_usage()

        st.text("Output")
        with st.spinner("Interpreting your text (This may take some time)"):
            if explanation_class_choice != "predicted":
                word_attributions = cls_explainer(
                    text,
                    class_name=explanation_class_choice,
                    embedding_type=emb_type_num,
                    internal_batch_size=2,
                )
            else:
                word_attributions = cls_explainer(
                    text, embedding_type=emb_type_num, internal_batch_size=2
                )

        if word_attributions:
            word_attributions_expander = st.beta_expander(
                "Click here for raw word attributions"
            )
            with word_attributions_expander:
                st.json(word_attributions)
            components.v1.html(
                cls_explainer.visualize()._repr_html_(), scrolling=True, height=350
            )


if __name__ == "__main__":
    main()