An example of a mechanism that returns a prediction by HTTP from the result of machine learning

Introduction

The other day, I created a program that predicts accounts using machine learning results, but there was an opinion that the response of the prediction results was abnormally slow and unusable, so I tried a little ingenuity.

Learning accounting data and predicting accounts from the contents of the description when entering journals --Qiita

Response mechanism

I decided to use HTTP in order to easily realize that the learning result is stored in the memory and the account is returned when the abstract is sent.

So, I created an HTTP server with Python, read the learning result when the HTTP server was started, and when I sent the summary with GET, I predicted and returned the account.

Build an HTTP server with Python

I referred to the following article.

Easily create an HTTP server with Python-Qiita

I'm using it as it is, but it has been corrected a little because the library called BaseHTTPServer has been changed and an error occurred in Japanese processing. By the way, it's Python3.

CallbackServer.py

#!/usr/local/bin/python
# coding: utf-8

import requests
import http.server
import socketserver
from http.server import BaseHTTPRequestHandler
from urllib.parse import urlparse, unquote

def start(port, callback):
    def handler(*args):
        CallbackServer(callback, *args)
    server = socketserver.TCPServer(('', int(port)), handler)
    server.serve_forever()

class CallbackServer(BaseHTTPRequestHandler):
    def __init__(self, callback, *args):
        self.callback = callback
        BaseHTTPRequestHandler.__init__(self, *args)

    def do_GET(self):
        parsed_path = urlparse(self.path)
        query = unquote(parsed_path.query)
        
        self.send_response(200)
        self.end_headers()
        
        result = self.callback(query)
        
        self.wfile.write(result.encode('utf-8'))
        
        return

Server startup script

It reads the learning result at startup and returns the result predicted by the GET callback.

server.py

#!/usr/local/bin/python
# coding: utf-8

import sys
import CallbackServer

import pandas as pd
import numpy as np

homedir = "/home/scripts/"
filename = "data/code.csv"

df = pd.read_csv(homedir + filename, header=None)
df.index = df.pop(0)

df_rs = df.pop(1)

from sklearn.externals import joblib

scaler = joblib.load(homedir + 'data/scaler.pkl')
clf = joblib.load(homedir + 'data/clf.pkl')
vect = joblib.load(homedir + 'data/vect.pkl')

from janome.tokenizer import Tokenizer

t = Tokenizer()

def callback_method(query):
    texts = [
        query,
    ]

    notes = []
    for note in texts:
        tokens = t.tokenize(note)
        words = ""
        for token in tokens:
            words += " " + token.surface
        notes.append(words)
    
    X = vect.transform(notes)
    
    result = clf.predict(X)
    ans = ""

    for i in range(len(texts)):
        ans = df_rs.ix[result[i]]
        
    return ans

if __name__ == '__main__':
    port = sys.argv[1]
    CallbackServer.start(port, callback_method)

Start with the following command.

python

$ chmod a+x server.py
$ ./server.py 8080 &

Operation test

Let's get the prediction result using Ruby.

test.rb

require 'net/http'
require 'uri'

puts Net::HTTP.get_print('localhost', URI.escape('/?Expressway usage fee'), 8080)

I will do it.

python

$ ruby test.rb
Travel expenses transportation

It was done (^-^)

Try incorporating it into a LINE bot.

Good feeling (^-^)

Please refer to the following article for how to make a LINE bot.

Create an autoresponder BOT with LINE's Messaging API

By the way, you can make friends with this LINE bot with the following QR code.

What should I do next?