problems with Statement Selection

Open nickums opened this issue 4 years ago • 0 comments

I have trained my chatbot solely with my own input, and specifying read_only = True to prevent new inputs being added. I am using my own 'get_best_response' algorithm. I am experiencing problems with input selection:- My own 'get_best_input_match' is not being called New input statements are sometimes included in the input matching and matched to themselves which results in a default/random response, which draws on a set of statements that I have NOT provided in training. Here is my code and output: jokes = [{"id":299,"type":"general","setup":"Where do hamburgers dance?", "punchline":"The meat-ball."}, {"id":330,"type":"general","setup":"Why did the belt go to prison?","punchline": "He held up a pair of pants!"}] for joke in jokes: print(f"setup:{joke['setup']} \n punchline:{joke['punchline']}")

from random import choice import chatterbot from chatterbot import ChatBot from chatterbot.comparisons import levenshtein_distance from chatterbot.comparisons import synset_distance from chatterbot.comparisons import jaccard_similarity import logging logging.basicConfig(level=logging.INFO) # Enable info level logging logger = logging.getLogger(name)

def get_best_input_match(s1,s2): # find which algo gives reasonable result S = synset_distance(s1,s2) L = synset_distance(s1,s2) J = synset_distance(s1,s2) logger.info(f'Syn,Lev,Jcc similarities are {S,L.J}') return max(S,max(L,J))

def get_best_response(input_statement, response_list, storage=None): i, best = 0, 0 for response in response_list: p = response.confidence if p > best: best, k = p, i i += 1 if best == 0: logger.info(f'Selecting no confidence random response.') return choice(response_list) # random response from the selection else: logger.info(f'Selecting response {k} with highest confidence {best}') return response_list[best]

def chat(input):
response = chatter.get_response(input) print(f'{input}\n{response.confidence}:{response.text}\n')

chatter = ChatBot("The Joker", read_only = True, # dont learn/add new stmts response_selection_method = get_best_response, statement_comparision_function = get_best_input_match)

from chatterbot.trainers import ListTrainer trainer = ListTrainer(chatter)

ONLY train the jokes - but there are a lot of statements already in the dB?

for item in jokes: joke = [item['setup'], item['punchline']]; print(joke) trainer.train(joke)

chat('Why did the belt go to prison?') # T 1 -> matches out of 11(!) optimal responses chat('why is the belt in prison?') # S1 .82 -> bestmatch Train stmt T -> OK chat('why was the belt imprisoned?') # S2 .72 matches itself and T -> OK chat('the belt was imprisoned for what offence?') # S3 searches T S1 S3 matches S3 -> random chat('why prosecute the belt?') # S4 .53 matches T to S4 -> OK

output: ! ['Where do hamburgers dance?', 'The meat-ball.']

List Trainer: [########## ] 50% List Trainer: [####################] 100%

['Why did the belt go to prison?', 'He held up a pair of pants!']

List Trainer: [########## ] 50% List Trainer: [####################] 100%

INFO:chatterbot.chatterbot:Beginning search for close text match INFO:chatterbot.chatterbot:Processing search results INFO:chatterbot.chatterbot:Similar text found: Someone said they're going to stop allowing machines in speedrun competitions. Search me why, they'd have to drop every tool-assisted speedrun out there. 0.22 INFO:chatterbot.chatterbot:Similar text found: i always say, if you see an ass go by, kiss it. 0.36 INFO:chatterbot.chatterbot:Similar text found: Why did the belt go to prison? 1.0 INFO:chatterbot.chatterbot:Using "Why did the belt go to prison?" as a close match to "Why did the belt go to prison?" with a confidence of 1.0 INFO:chatterbot.chatterbot:Selecting response from 11 optimal responses. INFO:main:Selecting no confidence random response. INFO:chatterbot.chatterbot:Response selected. Using "He held up a pair of pants!" INFO:chatterbot.chatterbot:BestMatch selected "He held up a pair of pants!" as a response with a confidence of 1.0 Why did the belt go to prison? 1.0:He held up a pair of pants!

INFO:chatterbot.chatterbot:Beginning search for close text match INFO:chatterbot.chatterbot:Processing search results INFO:chatterbot.chatterbot:Similar text found: Why did the belt go to prison? 0.82 INFO:chatterbot.chatterbot:Similar text found: why was the belt in prison? 0.94 INFO:chatterbot.chatterbot:Using "Why did the belt go to prison?" as a close match to "why is the belt in prison?" with a confidence of 0.82 INFO:chatterbot.chatterbot:Selecting response from 11 optimal responses. INFO:main:Selecting no confidence random response. INFO:chatterbot.chatterbot:Response selected. Using "He held up a pair of pants!" INFO:chatterbot.chatterbot:BestMatch selected "He held up a pair of pants!" as a response with a confidence of 0.82 why is the belt in prison? 0.82:He held up a pair of pants!

INFO:chatterbot.chatterbot:Beginning search for close text match INFO:chatterbot.chatterbot:Processing search results INFO:chatterbot.chatterbot:Similar text found: Why did the belt go to prison? 0.72 INFO:chatterbot.chatterbot:Similar text found: why was the belt in prison? 0.91 INFO:chatterbot.chatterbot:Using "Why did the belt go to prison?" as a close match to "why was the belt imprisoned?" with a confidence of 0.72 INFO:chatterbot.chatterbot:Selecting response from 11 optimal responses. INFO:main:Selecting no confidence random response. INFO:chatterbot.chatterbot:Response selected. Using "He held up a pair of pants!" INFO:chatterbot.chatterbot:BestMatch selected "He held up a pair of pants!" as a response with a confidence of 0.72 why was the belt imprisoned? 0.72:He held up a pair of pants!

INFO:chatterbot.chatterbot:Beginning search for close text match INFO:chatterbot.chatterbot:Processing search results INFO:chatterbot.chatterbot:Similar text found: Why did the belt go to prison? 0.48 INFO:chatterbot.chatterbot:Similar text found: why was the belt in prison? 0.5 INFO:chatterbot.chatterbot:Similar text found: the belt imprisoned for what offence? 0.95 INFO:chatterbot.chatterbot:Using "the belt imprisoned for what offence?" as a close match to "the belt was imprisoned for what offence?" with a confidence of 0.95 INFO:chatterbot.chatterbot:No responses found. Generating alternate response list. INFO:chatterbot.chatterbot:No known response to the input was found. Selecting a random response. INFO:chatterbot.chatterbot:BestMatch selected "Someone said they're going to stop allowing machines in speedrun competitions. Search me why, they'd have to drop every tool-assisted speedrun out there." as a response with a confidence of 0 the belt was imprisoned for what offence? 0:Someone said they're going to stop allowing machines in speedrun competitions. Search me why, they'd have to drop every tool-assisted speedrun out there.

Apr 18 '21 15:04 nickums