mirror of
https://github.com/fotoente/MIsskey-ebooks-bot
synced 2024-11-29 06:40:44 +00:00
375 lines
12 KiB
Python
375 lines
12 KiB
Python
import requests
|
|
import json
|
|
import os
|
|
import sys
|
|
import regex
|
|
import configparser
|
|
import markovify
|
|
import sqlite3
|
|
from pathlib import Path
|
|
from datetime import *
|
|
from time import sleep
|
|
|
|
def check_str_to_bool(text) -> bool:
|
|
if (text == "True" or text == "true" or text == "TRUE"):
|
|
return True
|
|
elif (text == "False" or text == "false" or text == "FALSE"):
|
|
return False
|
|
else:
|
|
return True
|
|
|
|
def get_notes(**kwargs):
|
|
noteid = "k"
|
|
sinceid = ""
|
|
min_notes = 0
|
|
notesList = []
|
|
returnList = []
|
|
|
|
if (kwargs):
|
|
if ("min_notes" in kwargs):
|
|
#print("min_notes found!")
|
|
init = True
|
|
min_notes = kwargs["min_notes"]
|
|
|
|
elif ("lastnote" in kwargs):
|
|
#print("Lastnote found!")
|
|
init = False
|
|
sinceid = kwargs["lastnote"]
|
|
|
|
else:
|
|
print("Wrong arguments given!")
|
|
print("Exiting routine!")
|
|
return
|
|
else:
|
|
print("No arguments given!")
|
|
print("Exiting routine")
|
|
return None
|
|
|
|
#Load configuration
|
|
config = configparser.ConfigParser()
|
|
config.read(os.path.join(os.path.dirname(__file__), 'bot.cfg'))
|
|
#print(os.path.join(os.path.dirname(__file__), 'bot.cfg'))
|
|
|
|
url="https://"+config.get("misskey","instance_read")+"/api/users/show"
|
|
host=config.get("misskey","instance_read")
|
|
try:
|
|
req = requests.post(url, json={"username" : config.get("misskey","user_read"), "host" : host})
|
|
req.raise_for_status()
|
|
except requests.exceptions.HTTPError as err:
|
|
print("Couldn't get Username! " + str(err))
|
|
sys.exit(1)
|
|
|
|
|
|
userid = req.json()["id"]
|
|
|
|
#Read & Sanitize Inputs from Config File
|
|
try:
|
|
includeReplies = check_str_to_bool(config.get("markov","includeReplies"))
|
|
except (TypeError, ValueError) as err:
|
|
includeReplies = True
|
|
|
|
try:
|
|
includeMyRenotes = check_str_to_bool(config.get("markov","includeMyRenotes"))
|
|
except (TypeError, ValueError) as err:
|
|
includeMyRenotes = False
|
|
|
|
try:
|
|
excludeNsfw = check_str_to_bool(config.get("markov","excludeNsfw"))
|
|
except (TypeError, ValueError) as err:
|
|
excludeNsfw = True
|
|
|
|
run = True
|
|
oldnote=""
|
|
|
|
while run:
|
|
|
|
if ((init and len(notesList) >= min_notes) or (oldnote == noteid)):
|
|
break
|
|
|
|
try:
|
|
req = requests.post("https://"+config.get("misskey","instance_read")+"/api/users/notes", json = {
|
|
"userId": userid,
|
|
"includeReplies" : includeReplies,
|
|
"limit" : 100,
|
|
"includeMyRenotes" : includeMyRenotes,
|
|
"withFiles" : False,
|
|
"excludeNsfw" : excludeNsfw,
|
|
"untilId" : noteid,
|
|
"sinceId" : sinceid
|
|
})
|
|
req.raise_for_status()
|
|
except requests.exceptions.HTTPError as err:
|
|
print("Couldn't get Posts! "+str(err))
|
|
sys.exit(1)
|
|
|
|
for jsonObj in req.json():
|
|
notesList.append(jsonObj)
|
|
if (len(notesList) == 0):
|
|
print("No new notes to load!")
|
|
return 0
|
|
|
|
oldnote = noteid
|
|
|
|
noteid = notesList[len(notesList)-1]["id"]
|
|
|
|
print(str(len(notesList)) + " Notes read.")
|
|
print("Processing notes...")
|
|
|
|
for element in notesList:
|
|
lastTime = element["createdAt"]
|
|
lastTimestamp = int(datetime.timestamp(datetime.strptime(lastTime, '%Y-%m-%dT%H:%M:%S.%f%z'))*1000)
|
|
|
|
content = element["text"]
|
|
|
|
if content is None: #Skips empty notes (I don't know how there could be empty notes)
|
|
continue
|
|
|
|
content = regex.sub(r"(?>@(?>[\w\-])+)(?>@(?>[\w\-\.])+)?", '', content) #Remove instance name with regular expression
|
|
content = content.replace("::",": :") #Break long emoji chains
|
|
content = content.replace("@", "@"+chr(8203))
|
|
|
|
dict = {"id" : element["id"], "text" : content, "timestamp" : lastTimestamp}
|
|
returnList.append(dict)
|
|
|
|
return returnList
|
|
|
|
def calculate_markov_chain():
|
|
text = ""
|
|
#Load configuration
|
|
config = configparser.ConfigParser()
|
|
config.read((Path(__file__).parent).joinpath('bot.cfg'))
|
|
try:
|
|
max_notes = config.get("markov","max_notes")
|
|
except (TypeError, ValueError) as err:
|
|
max_notes = "10000"
|
|
|
|
databasepath = (Path(__file__).parent).joinpath('roboduck.db')
|
|
if (not (os.path.exists(databasepath) and os.stat(databasepath).st_size != 0)):
|
|
print("Roboduck database not already created!")
|
|
print("Exit initialization!")
|
|
sys.exit(0)
|
|
|
|
with open(databasepath, 'r', encoding='utf-8') as emojilist:
|
|
database = sqlite3.connect(databasepath)
|
|
|
|
data = database.cursor()
|
|
data.execute("SELECT text FROM notes ORDER BY timestamp DESC LIMIT " + max_notes + ";")
|
|
|
|
rows = data.fetchall()
|
|
|
|
for row in rows:
|
|
text += row[0] + "\n"
|
|
|
|
markovchain = markovify.Text(text)
|
|
markovchain.compile(inplace = True)
|
|
|
|
markov_json = markovchain.to_json()
|
|
|
|
with open((Path(__file__).parent).joinpath('markov.json'), "w", encoding="utf-8") as markov:
|
|
json.dump(markov_json, markov)
|
|
|
|
def clean_database():
|
|
databasepath = (Path(__file__).parent).joinpath('roboduck.db')
|
|
if (not (os.path.exists(databasepath) and os.stat(databasepath).st_size != 0)):
|
|
print("No database found!")
|
|
print("Please run Bot first!")
|
|
sys.exit(0)
|
|
|
|
with open(databasepath, "a", encoding="utf-8") as f:
|
|
database = sqlite3.connect(databasepath)
|
|
|
|
#Reading config file bot.cfg with config parser
|
|
config = configparser.ConfigParser()
|
|
config.read((Path(__file__).parent).joinpath('bot.cfg'))
|
|
#print((Path(__file__).parent).joinpath('bot.cfg'))
|
|
try:
|
|
max_notes = config.get("markov","max_notes")
|
|
except (TypeError, ValueError) as err:
|
|
max_notes = "10000"
|
|
|
|
data = database.cursor()
|
|
data.execute("DELETE FROM notes WHERE id NOT IN (SELECT id FROM notes ORDER BY timestamp DESC LIMIT " + max_notes + ");")
|
|
|
|
database.commit()
|
|
database.close()
|
|
|
|
def create_sentence():
|
|
with open((os.path.join((Path(__file__).parent), 'markov.json')), "r", encoding="utf-8") as markov:
|
|
markov_json = json.load(markov)
|
|
|
|
text_model = markovify.Text.from_json(markov_json)
|
|
|
|
note=""
|
|
|
|
#Reading config file bot.cfg with config parser
|
|
config = configparser.ConfigParser()
|
|
config.read((Path(__file__).parent).joinpath('bot.cfg'))
|
|
#print((Path(__file__).parent).joinpath('bot.cfg'))
|
|
#Read & Sanitize Inputs
|
|
try:
|
|
test_output = check_str_to_bool(config.get("markov","test_output"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("test_output: " + str(err))
|
|
test_output = True
|
|
|
|
if (test_output):
|
|
try:
|
|
tries = int(config.get("markov","tries"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("tries: " + str(err))
|
|
tries = 250
|
|
|
|
try:
|
|
max_overlap_ratio = float(config.get("markov","max_overlap_ratio"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("max_overlap_ratio: " + str(err))
|
|
max_overlap_ratio = 0.7
|
|
|
|
try:
|
|
max_overlap_total = int(config.get("markov","max_overlap_total"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("max_overlap_total: " + str(err))
|
|
max_overlap_total = 10
|
|
|
|
try:
|
|
max_words = int(config.get("markov","max_words"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("max_words: " + str(err))
|
|
max_words = None
|
|
|
|
try:
|
|
min_words = int(config.get("markov","min_words"))
|
|
except (TypeError, ValueError) as err:
|
|
#print("min_words: " + str(err))
|
|
min_words = None
|
|
|
|
if (max_words is not None and min_words is not None):
|
|
if (min_words >= max_words):
|
|
#print("min_words ("+str(min_words)+") bigger than max_words ("+str(max_words)+")! Swapping values!")
|
|
swap = min_words
|
|
min_words = max_words
|
|
max_words = swap
|
|
|
|
else:
|
|
tries = 250
|
|
max_overlap_ratio = 0.7
|
|
max_overlap_total = 15
|
|
max_words = None
|
|
min_words = None
|
|
|
|
"""
|
|
#Debug section to rpint the used values
|
|
print("These values are used:")
|
|
print("test_output: " + str(test_output))
|
|
print("tries: " + str(tries))
|
|
print("max_overlap_ratio: " + str(max_overlap_ratio))
|
|
print("max_overlap_total: " + str(max_overlap_total))
|
|
print("max_words: " + str(max_words))
|
|
print("min_words: " + str(min_words))
|
|
"""
|
|
|
|
#Applying Inputs
|
|
note = text_model.make_sentence(
|
|
test_output = test_output,
|
|
tries = tries,
|
|
max_overlap_ratio = max_overlap_ratio,
|
|
max_overlap_total = max_overlap_total,
|
|
max_words = max_words,
|
|
min_words = min_words
|
|
)
|
|
if (note is not None):
|
|
return note
|
|
else:
|
|
return "Error in markov chain sentence creation: Couldn't calculate sentence!\n\n☹ Please try again! "
|
|
|
|
def update():
|
|
notesList = []
|
|
databasepath = (Path(__file__).parent).joinpath('roboduck.db')
|
|
if (not (os.path.exists(databasepath) and os.stat(databasepath).st_size != 0)):
|
|
print("No database found!")
|
|
print("Please run Bot first!")
|
|
sys.exit(0)
|
|
|
|
with open(databasepath, "a", encoding="utf-8") as f:
|
|
database = sqlite3.connect(databasepath)
|
|
print("Connected to roboduck.db succesfull...")
|
|
|
|
data = database.cursor()
|
|
data.execute("SELECT id FROM notes WHERE timestamp = (SELECT MAX(timestamp) FROM notes);")
|
|
|
|
sinceNote = data.fetchone()[0]
|
|
|
|
notesList = get_notes(lastnote = sinceNote)
|
|
|
|
if (notesList == 0):
|
|
database.close()
|
|
return
|
|
|
|
print("Insert new notes to database...")
|
|
for note in notesList:
|
|
database.execute("INSERT OR IGNORE INTO notes (id, text, timestamp) VALUES(?, ?, ?)", [note["id"], note["text"], note["timestamp"]])
|
|
|
|
database.commit()
|
|
print("Notes updated!")
|
|
database.close()
|
|
|
|
print("Cleaning database...")
|
|
clean_database()
|
|
print("Database cleaned!")
|
|
|
|
print("Short sleep to prevent file collison...")
|
|
sleep(10)
|
|
|
|
print("Calculating new Markov Chain...")
|
|
calculate_markov_chain()
|
|
print("Markov Chain saved!")
|
|
|
|
print("\nUpdate done!")
|
|
|
|
def init_bot():
|
|
notesList = []
|
|
databasepath = (Path(__file__).parent).joinpath('roboduck.db')
|
|
if (os.path.exists(databasepath) and os.stat(databasepath).st_size != 0):
|
|
print("Roboduck database already created!")
|
|
print("Exit initialization!")
|
|
sys.exit(0)
|
|
|
|
print("Creating database...")
|
|
|
|
with open(databasepath, "w+", encoding="utf-8") as f:
|
|
database = sqlite3.connect(databasepath)
|
|
print("Connected to roboduck.db succesfull...")
|
|
|
|
print("Creating Table...")
|
|
database.execute("CREATE TABLE notes (id CHAR(10) PRIMARY KEY, text CHAR(5000), timestamp INT);")
|
|
|
|
print("Table NOTES created...")
|
|
|
|
#Load configuration
|
|
config = configparser.ConfigParser()
|
|
config.read((Path(__file__).parent).joinpath('bot.cfg'))
|
|
try:
|
|
initnotes = int(config.get("markov","min_notes"))
|
|
except (TypeError, ValueError) as err:
|
|
#print(err)
|
|
initnotes=1000
|
|
|
|
print("Try reading first " + str(initnotes) + " notes.")
|
|
|
|
notesList = get_notes(min_notes = initnotes)
|
|
|
|
print("Writing notes into database...")
|
|
|
|
for note in notesList:
|
|
database.execute("INSERT INTO notes (id, text, timestamp) VALUES(?, ?, ?)", [note["id"], note["text"], note["timestamp"]])
|
|
|
|
database.commit()
|
|
database.close()
|
|
|
|
print("Notes written...")
|
|
print("Creating Markov Chain")
|
|
calculate_markov_chain()
|
|
|
|
print("Markov Chain calculated & saved.\n")
|
|
print("Finished initialization!\n")
|
|
print("The bot will now be started!")
|