app.py

import gradio as gr
import pypandoc
import glob
import shutil
import os
import tqdm
from huggingface_hub import snapshot_download
from huggingface_hub import HfApi
import tempfile
import re
from pdfminer.high_level import extract_text
import time

HF_TOKEN = os.environ.get("HF_TOKEN")

api = HfApi()


#from docx import Document
#document = Document()
#document.add_heading('Labels for ', level=1)
RESULTS_FOLDER = "./results"

DOC_FOLDER = snapshot_download("claudiag/atlas", token=HF_TOKEN, repo_type="dataset")

CAT_TO_CODEWORDS = {
    "Prejudices": ["prejudice", "judge", "preconceive", "stigma", "assumption", "assume", "misunderstanding", "unexamined", "distorted", "clear", "compar"],
    "Self-knowledge": ["self-knowledge", "self-awareness", "introspection", "examined", "myself", "realization", "belief"],
    "Similarities": ["similarity", "same", "similar", "equal", "related", "together"],
    "Diversity": ["diverse", "different", "diverse", "particular", "range", "multiplicity"],
    "Business school": ["ESADE", "competitive", "business school", "education", "study", "university", "student", "consulting", "professional", "pressure", "performance", "institution"],
    "Courage": ["courage", "brave", "dare", "step", "determine"],
    "Change": ["change", "finally", "at last", "decided", "chose", "concluded", "want to", "swap", "different", "not the same", "replace", "convert", "trade", "future", "decision"],
    "Coherence": ["coherent", "align", "incoherent", "consistent"],
    "Voicing": ["speak", "express", "voice", "talk", "say", "open up", "articulate", "communicate", "convey", "reveal", "show", "verbalize", "phrase", "word"],
    "Listening": ["listen", "pay attention", "quiet", "silence", "process", "hear", "attend"],
    "Understanding": ["learn", "understand", "realize", "see", "believe", "question", "critical", "thought", "reasonable", "logical", "rational", "comprehensible", "accept"],
    "Relationships": ["relationship", "relate", "bond", "connection", "bond", "others", "appreciate", "appreciation", "recognize", "recognition", "acknowledge"],
    "Emotions": ["emotions", "felt", "feel", "a feeling of", "sense", "sensation", "instinct", "sentiment", "gut feeling", "intense", "wave"],
    "The course": ["first time", "never", "always", "course", "elective", "Socratic Dialogue", "dialogue", "debate", "enroll", "arguments"],
}

CATEGORIES = CAT_TO_CODEWORDS.keys()

def retrieve_lines(filename):
    extension = filename.split(".")[-1]

    if extension == "pdf":
        text = extract_text(filename)
        lines = text.split("\n")
    elif extension in ["docx", "doc"]:
        with tempfile.TemporaryDirectory() as tmpdirname:
            outfile = os.path.join(tmpdirname, "temp.txt")
            pypandoc.convert_file(filename, 'plain', outputfile=outfile)
            with open(outfile, "r") as f:
                lines = f.readlines()
        
        lines = [l.strip() for l in lines]

    lines = " ".join(lines)
    lines = lines.split(".")

    return lines

def match_code(lines, codewords):
    match_dict = {}
    keywords_to_match = re.compile(fr'\b(?:{"|".join(codewords)})\b')
    for i, _ in enumerate(lines):
        line = lines[i]
        matches = list(keywords_to_match.finditer(line))

        if len(matches) > 0:
            for m in matches:
                span = m.span()
                line = line[:span[0]] + line[span[0]:span[1]].upper() + line[span[1]:]

            match_dict[i] = " ".join(line.rstrip().lstrip().split())

    return match_dict

def main(filename, codewords_mapping):
    lines = retrieve_lines(filename)
    files = []

    for label, codewords in codewords_mapping.items():
        match = match_code(lines, codewords)
        
        out = ""
        if len(match) > 0:
            result_file = ".".join(['_'.join(label.split()), "result", "txt"])
            result_file = os.path.join(RESULTS_FOLDER, result_file)
            if not os.path.exists(result_file):
                out += f"# Code: {label}\n"
                out += 25 * "="
                out += "\n\n"
    
            out += f"## Source: {'/'.join(filename.split('/')[-2:])}\n"
            out += 25 * "-"
            out += "\n"
            out += "\n".join([f'-{v}' for k,v in match.items()])
            out += "\n"
            out += 25 * "-"
            out += "\n\n"

            with open(result_file, "a") as f:
                f.write(out)

        files.append(result_file)

    return files


def convert(*keywords):
    codewords_mapping = {k: v for k,v in zip(CATEGORIES, keywords)}

    num_files = 0

    shutil.rmtree(RESULTS_FOLDER, ignore_errors=True)
    os.makedirs(RESULTS_FOLDER)

    result_files = []
    folders = glob.glob(os.path.join(DOC_FOLDER, "*"))


    for folder in tqdm.tqdm(folders):
        all_files = tqdm.tqdm(glob.glob(f"{folder}/*"))
        num_files += len(all_files)

        for filename in all_files:
            try:
                result_files += main(filename, codewords_mapping)
            except Exception as e:
                print(f"{filename} not working because \n {e}")

    result_files = list(set(result_files))

    api.upload_folder(
        repo_id="patrickvonplaten/atlas",
        folder_path=RESULTS_FOLDER,
        path_in_repo=f"results_{time.time()}",
        repo_type="dataset",
        token=HF_TOKEN,
    )

    return f"Done. Processed {num_files} files."


inputs = [gr.Textbox(label=f"Enter your keywords for {k}", max_lines=2, placeholder=CAT_TO_CODEWORDS[k], value=",".join(CAT_TO_CODEWORDS[k])) for k in CATEGORIES]

iface = gr.Interface(
        fn=convert, inputs=inputs, outputs="text")
iface.launch()