apt.py

from random import randrange, choice
import string
from time import time
from math import exp

from bleurt.score import BleurtScorer

# from simpletransformers.classification import ClassificationModel
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
from pandas import DataFrame
from numpy import argmax

from flask import Flask, request, jsonify, session, render_template
from flask_cors import CORS, cross_origin
from flask_session import Session
from waitress import serve

bleurt_scorer = BleurtScorer("/home/animesh/MIforSE/bleurt-score/bleurt/bleurt-base-128/")
hg_model_hub_name = "ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli"
tokenizer = AutoTokenizer.from_pretrained(hg_model_hub_name)
model = AutoModelForSequenceClassification.from_pretrained(hg_model_hub_name)  # predicts E, N, C
# mi_scorer = ClassificationModel('roberta', 'roberta_nli/', use_cuda=False, args = {'reprocess_input_data':True})


def get_mi_score(s1, s2):  # returns average of s1 and s2
    tokenized_input_seq_pair = tokenizer.encode_plus(s1, s2, max_length=256, return_token_type_ids=True, truncation=True)
    input_ids = torch.Tensor(tokenized_input_seq_pair["input_ids"]).long().unsqueeze(0)
    token_type_ids = torch.Tensor(tokenized_input_seq_pair["token_type_ids"]).long().unsqueeze(0)
    attention_mask = torch.Tensor(tokenized_input_seq_pair["attention_mask"]).long().unsqueeze(0)
    outputs = model(
        input_ids,
        attention_mask=attention_mask,
        token_type_ids=token_type_ids,
        labels=None,
    )
    predicted_probability_12 = torch.softmax(outputs[0], dim=1)[0].tolist()  # batch_size only one

    tokenized_input_seq_pair = tokenizer.encode_plus(s2, s1, max_length=256, return_token_type_ids=True, truncation=True)
    input_ids = torch.Tensor(tokenized_input_seq_pair["input_ids"]).long().unsqueeze(0)
    token_type_ids = torch.Tensor(tokenized_input_seq_pair["token_type_ids"]).long().unsqueeze(0)
    attention_mask = torch.Tensor(tokenized_input_seq_pair["attention_mask"]).long().unsqueeze(0)
    outputs = model(
        input_ids,
        attention_mask=attention_mask,
        token_type_ids=token_type_ids,
        labels=None,
    )
    predicted_probability_21 = torch.softmax(outputs[0], dim=1)[0].tolist()  # batch_size only one

    return int(argmax(predicted_probability_12) == 0 and argmax(predicted_probability_21) == 0)

    # _, s1s2, __ = mi_scorer.eval_model(DataFrame({'text_a':s1, 'text_b':s2, 'labels':2}))
    # _, s2s1, __ = mi_scorer.eval_model(DataFrame({'text_a':s2, 'text_b':s1, 'labels':2}))
    # print(s1s2[0], s2s1[0], argmax(s1s2[0]), argmax(s2s1[0]))
    # return int(s1s2[0][2] > 0 and argmax(s1s2[0]) == 2 and s2s1[0][2] > 0 and argmax(s2s1[0]) == 2)


mrpc = []  # [quality, id1, id2, s1, s2]
with open("mrpc/msr_paraphrase_train.txt", "r") as f:
    for l in f.readlines()[1:]:
        mrpc.append(l.split("\t"))

ppnmt = []  # [quality, id1, id2, s1, s2]
with open("ppnmt/czeng_test_engeng.txt", "r") as f:
    for l in f.readlines()[1:]:
        ppnmt.append(l.split("\t"))

app = Flask(__name__)
app.config["SESSION_TYPE"] = "redis"
cors = CORS(app)
app.config["CORS_HEADERS"] = "Content-Type"
app.config["DEBUG"] = True
app.config["SECRET_KEY"] = b"\xb3\xb6\x02\x08E\\\xcb.\x13b(\x0f\xfb\x15\xcf\xc5"
Session(app)

letters_digits = string.ascii_uppercase + string.digits


@app.route("/", methods=["GET", "POST"])
@cross_origin()
def init():
    session["token"] = "".join((choice(letters_digits) for i in range(10)))
    session["final_amt"] = 0.0
    session["sentence"] = None
    print(session["token"])
    print(session["final_amt"])
    return start()


@app.route("/start", methods=["GET", "POST"])
@cross_origin()
def start():
    if session["final_amt"] >= 20:
        return end()
    print("in start")
    session["dataset"] = choice(["mrpc", "ppnmt"])
    if session["dataset"] == "mrpc":
        session["sentence_index"] = randrange(len(mrpc))
        session["sentence"] = str(mrpc[session["sentence_index"]][3])
    else:
        session["sentence_index"] = randrange(len(ppnmt))
        session["sentence"] = str(ppnmt[session["sentence_index"]][1])
    print(session["sentence"])
    print(session["token"])
    print(session["final_amt"])
    session["start_time"] = time()
    return render_template("form.html", data=session)


@app.route("/check", methods=["POST"])
@cross_origin()
def check_candidate():
    if session["final_amt"] >= 20:
        return end()
    session["candidate"] = request.form.get("candidate").strip().replace("\n", " ")
    print(session["token"])
    print("Candidate:", str(session["candidate"]))
    if session["sentence"] in session["candidate"]:
        session["miscore"] = 1
        session["dollars"] = 0
    else:
        bleurtscore = (bleurt_scorer.score([session["sentence"]], [session["candidate"]])[0] + bleurt_scorer.score([session["candidate"]], [session["sentence"]])[0]) / 2
        print("BLEURT:", str(bleurtscore))
        session["miscore"] = get_mi_score(session["sentence"], session["candidate"])
        print("MI:", str(session["miscore"]))
        # session['dollars'] = round(max(0, (miscore - (1 / (1 + exp(-bleurtscore)))) / 2), 2)
        session["dollars"] = round(session["miscore"] / ((1 + exp(5 * bleurtscore)) ** 2), 2)
        print("Dollars:", str(session["dollars"]))
        with open("sentences/checks", "a+") as f:
            f.write(
                "\t".join(
                    [
                        session["token"],
                        str(time()),
                        str(time() - session["start_time"]),
                        session["dataset"],
                        str(session["sentence_index"]),
                        session["sentence"],
                        session["candidate"],
                        str(bleurtscore),
                        str(session["miscore"]),
                        str(session["dollars"]),
                    ]
                )
                + "\n"
            )
    return dict(session)


@app.route("/submit", methods=["POST"])
@cross_origin()
def submit_candidate():
    if session["final_amt"] >= 20:
        return end()
    session["candidate"] = request.form.get("candidate").strip().replace("\n", " ")
    print(session["token"])
    print("Candidate:", str(session["candidate"]))
    print("Sentence:", str(session["sentence"]))
    if session["sentence"] in session["candidate"]:
        session["dollars"] = 0
    else:
        if session["dataset"] == "mrpc":
            del mrpc[session["sentence_index"]]
        else:
            del ppnmt[session["sentence_index"]]
        bleurtscore = (bleurt_scorer.score([session["sentence"]], [session["candidate"]])[0] + bleurt_scorer.score([session["candidate"]], [session["sentence"]])[0]) / 2
        session["miscore"] = get_mi_score(session["sentence"], session["candidate"])
        # session['dollars'] = round(max(0, (miscore - (1 / (1 + exp(-bleurtscore)))) / 2), 2)
        session["dollars"] = round(session["miscore"] / ((1 + exp(5 * bleurtscore)) ** 2), 2)
        with open("sentences/submits", "a+") as f:
            f.write(
                "\t".join(
                    [
                        session["token"],
                        str(time()),
                        str(time() - session["start_time"]),
                        session["dataset"],
                        str(session["sentence_index"]),
                        session["sentence"],
                        session["candidate"],
                        str(bleurtscore),
                        str(session["miscore"]),
                        str(session["dollars"]),
                    ]
                )
                + "\n"
            )
        session["final_amt"] += session["dollars"]
    if session["final_amt"] >= 20:
        return end()
    return start()


@app.route("/end", methods=["GET", "POST"])
@cross_origin()
def end():
    print("in end")
    print(session["token"])
    print(session["final_amt"])
    with open("sentences/ends", "a+") as f:
        f.write("\t".join([session["token"], str(session["final_amt"])]) + "\n")
    if session["final_amt"] < 2:
        session["final_amt"] = 0
        session["token"] = "-"
    return render_template("end.html", data=session)


# app.run(host='0.0.0.0')
serve(app, host="0.0.0.0", port=5000)