Asking questions & queuing answers

1. Asking questions & queuing answers

Now we're going to explore some other tricks you can use to build up conversation flows. As your bot becomes more sophisticated, your state machine will gain more and more states, and ever more rules for transitioning between them. You should always try to reduce this complexity, and in these exercises you will learn some tricks for doing so.

2. Reusable patterns

Let's say also start selling coffee filters through your bot-powered coffee business. The filters come in two colors, white and brown. Let's say your supplier is a little unreliable, and sometimes your run out of stock. These two very similar conversations are then possible. One way to handle these cases is to add an additional state which says that an alternative has been offered to the user, and adding some policy rules to handle the "yes" and "no" answers. But asking your user for confirmation before proceeding with something is a common pattern, and adding more states increases the complexity of your code. You can avoid this complexity by saving the follow-up as a "pending action"

3. Pending actions

To implement this, create a policy which returns two values the selected next `action` and a `pending_action`. The pending action will in many cases be `None`. Save the pending action in the outer scope, where the params dictionary and the state variable were also defined. - if we get a "yes" intent and there is a pending action, we execute it - if we get a "no" intent, we wipe any pending actions

4. Pending state transitions

In addition to pending actions, saving pending state transitions is another trick for keeping your state space small & complexity low. In this example, need to request some authentication info from the user First the users tells us they'd like to order coffee. We are in the init state, the selection action is to request authorization info, and we have a pending transition to the authed state. We request the user's phone number, and they supply it. Now, we immediately transition to the authed state, and we're ready for the user to place an order.

5. Let's practice!

Now it's your turn to try these techniques in some realistic bot scenarios.

This exercise is part of the course

Building Chatbots in Python

IntermediateSkill Level

4.7+

Start Course for Free

In this chapter, you'll learn how to build your first chatbot. After gaining a bit of historical context, you'll set up a basic structure for receiving text and responding to users, and then learn how to add the basic elements of personality. You'll then build rule-based systems for parsing text.

Exercise 1: Introduction to conversational software Exercise 2: EchoBot I Exercise 3: EchoBot II Exercise 4: Creating a personality Exercise 5: Chitchat Exercise 6: Adding variety Exercise 7: ELIZA I: asking questions Exercise 8: Text processing with regular expressions Exercise 9: ELIZA II: Extracting key phrases Exercise 10: ELIZA III: Pronouns Exercise 11: ELIZA IV: Putting it all together

Here, you'll use machine learning to turn natural language into structured data using spaCy, scikit-learn, and rasa NLU. You'll start with a refresher on the theoretical foundations and then move onto building models using the ATIS dataset, which contains thousands of sentences from real people interacting with a flight booking system.

Exercise 1: Understanding intents and entities Exercise 2: Intent classification with regex I Exercise 3: Intent classification with regex II Exercise 4: Entity extraction with regex Exercise 5: Word vectors Exercise 6: word vectors with spaCy Exercise 7: Intents and classification Exercise 8: Intent classification with sklearn Exercise 9: Entity extraction Exercise 10: Using spaCy's entity recognizer Exercise 11: Assigning roles using spaCy's parser Exercise 12: Robust language understanding with rasa NLU Exercise 13: Rasa NLU Exercise 14: Data-efficient entity recognition

In this chapter, you'll build a personal assistant to help you plan a trip. It will be able to respond to questions like "are there any cheap hotels in the north of town?" by looking inside a hotel’s database for matching results.

Exercise 1: Virtual assistants and accessing data Exercise 2: SQL basics Exercise 3: SQL statements in Python Exercise 4: Exploring a DB with natural language Exercise 5: Creating queries from parameters Exercise 6: Using your custom function to find hotels Exercise 7: Creating SQL from natural language Exercise 8: Incremental slot filling and negation Exercise 9: Refining your search Exercise 10: Basic negation Exercise 11: Filtering with excluded slots

Everything you've built so far has statelessly mapped intents to actions and responses. It's amazing how far you can get with that! But to build more sophisticated bots you will always want to add some statefulness. That's what you'll do here, as you build a chatbot that helps users order coffee.

Exercise 1: Why statefulness is key Exercise 2: Form filling Exercise 3: Asking contextual questions Exercise 4: Dealing with rejection Exercise 5: Asking questions & queuing answers

Current Exercise

Exercise 6: Pending actions I Exercise 7: Pending actions II Exercise 8: Pending state transitions Exercise 9: Putting it all together I Exercise 10: Putting it all together II Exercise 11: Frontiers of dialogue research Exercise 12: Generating text with neural networks Exercise 13: Congratulations!