slack_overflow.py

from flask import Flask
import os
import logging
from joblib import dump, load
from flask import request, jsonify, abort
from slack_sdk import WebClient
from slack_sdk.errors import SlackApiError
import json


from question_tagger import QuestionTagger
import torch
import numpy as np
from transformers import BertTokenizerFast as BertTokenizer, BertModel, AdamW, get_linear_schedule_with_warmup

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

client = WebClient(token=SLACK_TOKEN)
logger = logging.getLogger(__name__)
app = Flask(__name__)
# model = load("naivebayes.joblib")

BERT_MODEL_NAME = 'bert-base-cased'
bert = QuestionTagger(n_classes=3)
bert.load_state_dict(torch.load("trained_model.pth"))

channel_public_ids = {"python": "C028ZTHBP33", "cloud": "C0286GGP3PH", "node": "C028ZTLBQL9"}

@app.route('/', methods=["POST"])
def action():

	try:
		payload = json.loads(request.form["payload"])
		print(payload)
		actions = payload.get("actions")[0]
		action_id = actions.get("action_id")

		if action_id == "post":
			user = payload.get("user").get("username")
			value = json.loads(actions["value"])
			question = value["question"]
			channel_id = value["channel_id"]

			text = f"""New question from <@{user}>: 

{question}
			"""
			client.chat_postMessage(
				channel=channel_id,
				text=text
			)

			return jsonify({})

		elif action_id == "post_anon":
			value = json.loads(actions["value"])
			question = value["question"]
			channel_id = value["channel_id"]

			text = question
			client.chat_postMessage(
				channel=channel_id,
				text=text
			)

			return jsonify({})

	except SlackApiError as e:
		# print(f"Error: {e}")
		raise e


@app.route('/where', methods=['POST'])
def recommend():
    # Parse the parameters you need
	text = request.form.get('text', None)

	if text == None or text == '':
		return "Please enter a question."

	channel = predict(text)
	channel_id = channel_public_ids.get(channel)
	val = { "question" : text, "channel_id" : channel_id}

	return jsonify({
			"blocks": [
				{
					"type": "section",
					"text": {
						"type": "mrkdwn",
						"text": "*"+text+"*"
					}
				},
				{
					"type": "section",
					"text": {
						"type": "mrkdwn",
						"text": ":slack: Post your question in <#"+channel_id+"|"+channel+"> :slack:."
					}
				},
				{
					"type": "actions",
					"block_id": "actionblock789",
					"elements": [
						{
							"type": "button",
							"text": {
								"type": "plain_text",
								"text": "Post Question"
							},
							"style": "primary",
							"value": json.dumps(val),
							"action_id": "post"
						}, 
						{
							"type": "button",
							"text": {
								"type": "plain_text",
								"text": "Post Anonymously"
							},
							"style": "primary",
							"value": json.dumps(val),
							"action_id": "post_anon"
						}
					]
				}
			]
		})

def predict(text):
	test_comment = text
	tokenizer = BertTokenizer.from_pretrained(BERT_MODEL_NAME)
	encoding = tokenizer.encode_plus(
	  test_comment,
	  add_special_tokens=True,
	  max_length=128,
	  return_token_type_ids=False,
	  padding="max_length",
	  return_attention_mask=True,
	  return_tensors='pt',
	)

	_, test_prediction = bert(encoding["input_ids"], encoding["attention_mask"])
	test_prediction = test_prediction.flatten().detach().numpy()
	LABEL_COLUMNS = ['cloud', 'node', 'python']
	return LABEL_COLUMNS[np.argmax(test_prediction)]


if __name__ == '__main__':
    port = int(os.environ.get('PORT', 5000))
    app.run(host='0.0.0.0', port=port, debug=True)