app.py

import fasttext
import gradio as gr
from gradio import components as components
import numpy as np

model = "cc.en.300.bin"
model = fasttext.load_model(model)

def get_cosine(a, b):
    cosine_similarity = np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
    return cosine_similarity


def get_score(targets, answers, th):
    th = float(th)

    results = []
    targets = targets.split(",")
    target_len = len(targets)
    answers = answers.split(",")

    # th = 0.6
    for i, target in enumerate(targets):
        if target[0]==" ": target=target[1:]
        # if i>2:break
        target_vector = model.get_word_vector(target.lower())

        cosine_list = []
        # compare all of answer data
        for answer in answers:
            answer_vector = model.get_word_vector(answer.lower())
            cosine = get_cosine(target_vector, answer_vector)
            cosine_list.append(cosine)

        # comparison is over threshold -> true else -> false
        # results.append(max(cosine_list)>th)

        # reject answer once it is used
        if max(cosine_list)>th:
            argmax = cosine_list.index(max(cosine_list))
            results.append(max(cosine_list))
            answers.remove(answers[argmax])

        if len(answers)==0: break
    if len(results)==0: return 0
    result = int(sum(results) / len(results) * 100)
    # result = sum(results) / target_len
    return result


def my_app(targets, answers, th):
    return get_score(targets, answers, th)

inputs = [
    components.Textbox(label="Targets"),
    components.Textbox(label="Answers"),
    components.Textbox(label="Threshold", value=0.6),
]
output = components.Textbox(label="Output")

app = gr.Interface(
    fn=my_app, 
    inputs=inputs,
    outputs=output
    )
app.launch(
    app_kwargs={"docs_url":"/docs"}
)