Пример #1
0
    def __init__(self):
        Process.__init__(self)
        my_args = parse_args()

        my_args.changes = "indx_word = './model/ivocab.in.pkl', \
                    indx_word_target = './model/ivocab.out.pkl', \
                    word_indx = './model/vocab.in.pkl', \
                    word_indx_trgt = './model/vocab.out.pkl'"

        self.sampler = ChatSampler(my_args)
        self.RESPONSE_MAX_COST = 1.5
        self.NUM_SAMPLES = 10
        self.tokenizer_cmd = [
            os.getcwd() + '/tokenizer.perl', '-l', 'en', '-q', '-'
        ]
        self.detokenizer_cmd = [
            os.getcwd() + '/detokenizer.perl', '-l', 'fr', '-q', '-'
        ]
Пример #2
0
    def __init__(self):
        Process.__init__(self)
        my_args = parse_args()

        my_args.changes = "indx_word = './model/ivocab.in.pkl', \
                    indx_word_target = './model/ivocab.out.pkl', \
                    word_indx = './model/vocab.in.pkl', \
                    word_indx_trgt = './model/vocab.out.pkl'"
        self.sampler = ChatSampler(my_args)
        self.RESPONSE_MAX_COST = 1.5
        self.NUM_SAMPLES = 10
        self.tokenizer_cmd = [os.getcwd() + "/tokenizer.perl", "-l", "en", "-q", "-"]
        self.detokenizer_cmd = [os.getcwd() + "/detokenizer.perl", "-l", "fr", "-q", "-"]
Пример #3
0
    def __init__(self):
        Process.__init__(self)
        self.samplers = {}
        for l in conf.LANGUAGES:
            my_args = parse_args()
            my_args = conf.get_config(l, my_args)
            print(my_args)
            self.samplers[l] = ChatSampler(my_args)

        self.tokenizer_cmd = [
            os.getcwd() + '/tokenizer.perl', '-l', 'en', '-q', '-'
        ]
        self.detokenizer_cmd = [
            os.getcwd() + '/detokenizer.perl', '-l', 'fr', '-q', '-'
        ]
Пример #4
0
class ChatServer(Process):
    def __init__(self):
        Process.__init__(self)
        my_args = parse_args()

        my_args.changes = "indx_word = './model/ivocab.in.pkl', \
                    indx_word_target = './model/ivocab.out.pkl', \
                    word_indx = './model/vocab.in.pkl', \
                    word_indx_trgt = './model/vocab.out.pkl'"
        self.sampler = ChatSampler(my_args)
        self.RESPONSE_MAX_COST = 1.5
        self.NUM_SAMPLES = 10
        self.tokenizer_cmd = [os.getcwd() + "/tokenizer.perl", "-l", "en", "-q", "-"]
        self.detokenizer_cmd = [os.getcwd() + "/detokenizer.perl", "-l", "fr", "-q", "-"]

    def reverse(self, input_chat):
        return " ".join(input_chat.split(" ")[::-1])

    def sample_chat(self, input_chat):
        seq = self.reverse(input_chat)
        return self.sampler.sample(seq, self.NUM_SAMPLES)

    def filter_messages(self, messages, costs):
        return [(m, c) for (m, c) in zip(messages, costs) if (not m.startswith("khangbot")) and ("UNK" not in m)]

    def run(self):
        app = Flask(__name__)

        @app.route("/")
        def index():
            return "LSTM-based chat server"

        @app.route("/response_cost", methods=["GET", "POST", "PUT"])
        def response_cost_flask():
            if "val" in request.args:
                self.RESPONSE_MAX_COST = float(request.args.get("val"))
            return "Current max cost: %f" % self.RESPONSE_MAX_COST

        @app.route("/chat", methods=["GET", "POST", "PUT"])
        def chat_flask():
            debugging = "debug" in request.args
            sentence = request.args.get("input")

            print "Sentence: ", sentence
            try:
                tokenizer = subprocess.Popen(self.tokenizer_cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
                sentence, _ = tokenizer.communicate(sentence.encode("utf-8"))
            except UnicodeDecodeError:
                print "Got UnicodeDecodeError with sentence =", sentence
                return ""
            print "Sentence after tokenize: ", sentence

            [messages, costs] = self.sample_chat(sentence)

            if costs is not None and len(costs) > 0:
                messages = self.filter_messages(messages, costs)
                if len(messages) > 0:
                    decoded_messages = []
                    for (m, c) in messages:
                        try:
                            detokenizer = subprocess.Popen(
                                self.detokenizer_cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE
                            )
                            detokenized_sentence, _ = detokenizer.communicate(m.encode("utf-8"))
                            decoded_messages.append((m, c))
                        except UnicodeDecodeError:
                            print "Got UnicodeDecodeError in detokenizer with sentence =", m
                    if debugging:
                        return "\n".join(["%f\t%s" % (c, m) for (m, c) in decoded_messages])
                    return decoded_messages[0][0] if decoded_messages[0][1] < self.RESPONSE_MAX_COST else ""
            return ""

        @app.route("/shutdown", methods=["GET", "POST", "PUT"])
        def terminate_flask():
            func = request.environ.get("werkzeug.server.shutdown")
            if func is None:
                raise RuntimeError("Not running with the Werkzeug Server")
            func()
            return "OK"

        print "Listening to 0.0.0.0:5000..."
        app.run(host="0.0.0.0", debug=True, threaded=True, use_reloader=False)
Пример #5
0
class ChatServer(Process):
    def __init__(self):
        Process.__init__(self)
        my_args = parse_args()

        my_args.changes = "indx_word = './model/ivocab.in.pkl', \
                    indx_word_target = './model/ivocab.out.pkl', \
                    word_indx = './model/vocab.in.pkl', \
                    word_indx_trgt = './model/vocab.out.pkl'"

        self.sampler = ChatSampler(my_args)
        self.RESPONSE_MAX_COST = 1.5
        self.NUM_SAMPLES = 10
        self.tokenizer_cmd = [
            os.getcwd() + '/tokenizer.perl', '-l', 'en', '-q', '-'
        ]
        self.detokenizer_cmd = [
            os.getcwd() + '/detokenizer.perl', '-l', 'fr', '-q', '-'
        ]

    def reverse(self, input_chat):
        return ' '.join(input_chat.split(' ')[::-1])

    def sample_chat(self, input_chat):
        seq = self.reverse(input_chat)
        return self.sampler.sample(seq, self.NUM_SAMPLES)

    def filter_messages(self, messages, costs):
        return [(m, c) for (m, c) in zip(messages, costs)
                if (not m.startswith('khangbot')) and ('UNK' not in m)]

    def run(self):
        app = Flask(__name__)

        @app.route('/')
        def index():
            return 'LSTM-based chat server'

        @app.route('/response_cost', methods=['GET', 'POST', 'PUT'])
        def response_cost_flask():
            if 'val' in request.args:
                self.RESPONSE_MAX_COST = float(request.args.get('val'))
            return 'Current max cost: %f' % self.RESPONSE_MAX_COST

        @app.route('/chat', methods=['GET', 'POST', 'PUT'])
        def chat_flask():
            debugging = 'debug' in request.args
            sentence = request.args.get('input')

            print 'Sentence: ', sentence
            try:
                tokenizer = subprocess.Popen(self.tokenizer_cmd,
                                             stdin=subprocess.PIPE,
                                             stdout=subprocess.PIPE)
                sentence, _ = tokenizer.communicate(sentence.encode('utf-8'))
            except UnicodeDecodeError:
                print 'Got UnicodeDecodeError with sentence =', sentence
                return ""
            print 'Sentence after tokenize: ', sentence

            [messages, costs] = self.sample_chat(sentence)

            if costs is not None and len(costs) > 0:
                messages = self.filter_messages(messages, costs)
                if len(messages) > 0:
                    decoded_messages = []
                    for (m, c) in messages:
                        try:
                            detokenizer = subprocess.Popen(
                                self.detokenizer_cmd,
                                stdin=subprocess.PIPE,
                                stdout=subprocess.PIPE)
                            detokenized_sentence, _ = detokenizer.communicate(
                                m.encode('utf-8'))
                            decoded_messages.append((m, c))
                        except UnicodeDecodeError:
                            print 'Got UnicodeDecodeError in detokenizer with sentence =', m
                    if debugging:
                        return '\n'.join(
                            ['%f\t%s' % (c, m) for (m, c) in decoded_messages])
                    return decoded_messages[0][0] if decoded_messages[0][
                        1] < self.RESPONSE_MAX_COST else ""
            return ""

        @app.route('/shutdown', methods=['GET', 'POST', 'PUT'])
        def terminate_flask():
            func = request.environ.get('werkzeug.server.shutdown')
            if func is None:
                raise RuntimeError('Not running with the Werkzeug Server')
            func()
            return 'OK'

        print 'Listening to 0.0.0.0:5000...'
        app.run(host='0.0.0.0', debug=True, threaded=True, use_reloader=False)