GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → bixbydevelopers → capsule-sample-space-resorts

bixbydevelopers / capsule-sample-space-resorts

Licence: Apache-2.0 license
The Space Resorts sample capsule demonstrates a fully-developed capsule that uses several of the features and abilities discussed in the Bixby Developer Guides, while also using the best practices described in the Bixby Design Guides.

Programming Languages

javascript
184084 projects - #8 most used programming language

Labels

searchend-to-end-learningbixby

Projects that are alternatives of or similar to capsule-sample-space-resorts

capsule-sample-shoe
This sample capsule shows how to include basic search in a capsule. The user can start with a simple utterance, such as "Find me some shoes," or do a more specific search, such as "Find me dancing shoes."
Stars: ✭ 13 (-45.83%)
Mutual labels:  bixby
end-to-end-deep-learning
Autonomous driving simulation in the Unity engine.
Stars: ✭ 27 (+12.5%)
Mutual labels:  end-to-end-learning
weasel
Weakly Supervised End-to-End Learning (NeurIPS 2021)
Stars: ✭ 117 (+387.5%)
Mutual labels:  end-to-end-learning
Pontryagin-Differentiable-Programming
A unified end-to-end learning and control framework that is able to learn a (neural) control objective function, dynamics equation, control policy, or/and optimal trajectory in a control system.
Stars: ✭ 111 (+362.5%)
Mutual labels:  end-to-end-learning
Writing-editing-Network
Code for Paper Abstract Writing through Editing Mechanism
Stars: ✭ 72 (+200%)
Mutual labels:  end-to-end-learning
QuantumForest
Fast Differentiable Forest lib with the advantages of both decision trees and neural networks
Stars: ✭ 63 (+162.5%)
Mutual labels:  end-to-end-learning
bixby
we make your Artweek. we find your Inspiration. 💫
Stars: ✭ 20 (-16.67%)
Mutual labels:  bixby


Bixby Space Resorts Sample Capsule

Overview

This capsule is the the companion code to the Bixby Space Resorts Sample Capsule guide. Space Resorts is a fun capsule that allows you to book a space vacation! You can find a space resort, book a reservation, look up a reservation, and change or cancel a reservation. This advanced capsule demonstrates development and design of searches, transactions and UI.

Use cases

Finding Space Resorts

Outer "find" Queries

Outer "find" Query

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:SpaceResort#all -has:continue. Examples:

  • Find space resorts

  • Show me space hotels with crater canyoneering

  • Search for hotels around The Red Planet

  • Look for space hotels with quantum bungee jumping around Saturn

We train these to have the goal SpaceResort#all and annotate any resort names, planets and search criteria as Values. Here SpaceResort#all is a property projection to the all property of the SpaceResort. This property is a boolean of type ViewAll, which is always true and is used as a proxy to signify that we want to display the full space resort, instead of focusing on a single property like gravity or planet (see property projections below). A match-pattern ties the views and dialogs for this property projection to describe the whole hotel.

match {
  ViewAll(all) {
    from-property {
      SpaceResort (result)
    }
  }
}

Why treat this as a property projection instead of setting the goal to the SpaceResort Structure?

When the user wants to book a space resort and there are many possible candidates in context, the user picks a single one using a selection prompt, in order to continue with the booking (see SpaceResort Selection Prompt selection below for full details). The context for that prompt is SpaceResort, and selection prompt training must always have the same goal as its context. Accordingly, selection prompt training for SpaceResort needs to use SpaceResort as its goal.

The selection prompt training also uses a special flagged signal to route the plan through the SelectResort action that filters the hotels currently in context based on the newly provided inputs. It is crucial to add the SelectResort flagged signal to the selection prompt training annotations to achieve this behavior. However, we do not want to add the SelectResort flagged signal to the "find" queries because these should issue a new search via the FindSpaceResorts action instead of filtering existing results via the SelectResort action. Since the selection prompt requires different annotations patterns compared to the "find" queries, and annotation patterns must be consistent for the same goal, this means that they must use a different goal.

This example demonstrates two points. First, the selection prompt training must use SpaceResort as a goal. Second, that the "find" queries must use a different goal than the selection prompts. By putting these together, we deduce that the "find" queries cannot use SpaceResort as a goal. Therefore, we use a distinct goal (SpaceResort#all) for the "find" queries in order to provide consistent annotation patterns per goal.

Why don't we set the goal to the FindSpaceResorts action?

Another alternative would be to set the goal to the specific FindSpaceResorts action, making it very clear how to fulfill the request. This approach would simplify the match patterns for views, so we would not need to use the from-property key. However, we use the property projection approach so that our final resting context for "find" queries is the same as for property projection queries. This means that we can pivot between all these states seamlessly, or launch the "book" flow from any of these. Example conversation:

  1. Find space resorts

  2. The second one

  3. What planet is it on?

  4. What's the gravity there?

  5. Book it

Inner "find" Queries (Continuations)

Inner "find" Query

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:SpaceResort#all has:continue. Examples:

  • On Jupiter

  • With low gravity

  • Only the ones that are kid-friendly

These are continuations of the outer "find" queries that allow users to refine their space resorts search by providing additional inputs. Since the goal for outer "find" queries is SpaceResort#all, we annotate both the goal and the "Continuation of" to also be SpaceResort#all. Any resort names, planets and search criteria are annotated as Values. This reissues a search with the new inputs being added to those already in context.

Property Projection Flows (Planet, Gravity)

Outer Property Projections

Outer Property Projection

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:SpaceResort#* -goal:SpaceResort#all -has:continue. For example:

  • What's the gravity at The Mercurial?

  • Where is Io-Tel?

Here, the user is asking to know about a specific property of a space resort, such as the gravity or the planet. We train the goal to be that property projection (ex: SpaceResort#gravity), and we annotate any resort names, planets and search criteria as Values. We also add a special flagged signal route to ProjectResort. This is in case there were multiple space resorts that matched the search inputs. Then the ProjectResort action will ask the user to select a single space resort before providing the answer.

Inner Property Projections (Continuations)

Inner Property Projection

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:SpaceResort#* -goal:SpaceResort#all has:continue. For example:

  • What's the gravity there?

  • What planet is it on?

We train these just like the outer property projections, with the addition of a "Continuation of" SpaceResort. This allows pivoting between inner/outer "find" queries and inner/outer property projections.

Booking Space Resorts

Outer "book" Queries

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:MakeReservation -has:continue. For example:

  • Make a reservation for a space resort on Mars the third weekend in December for 2 astronauts

We train these to the goal MakeReservation, which is the Action to finalize the transaction. We also add two flagged signal routes: CreateItem and CreateOrder. We annotate as Values any present inputs for either "find" or "book", such as resort name, planet, search criteria, number of astronauts, etc. This creates a plan to first find a space resort that matches the search inputs, then prepare an Order and pass it to the MakeReservation action, which will present the user with a Confirmation screen to review and agree to the reservation.

Inner "book" Queries (Continuations of SpaceResort)

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:MakeReservation continuation:SpaceResort. For example:

  • Make reservation

  • Book a pod for 2 astronauts

We train these just like the outer "book" queries, with the addition of a "Continuation of" SpaceResort. This is to cover cases where the users are already browsing space resorts and want to initiate a booking for one of the results in context.

Inner "Change Order" Queries (Continuations to change the Order)

You can see all trained utterances and plans by entering this query in the training tab search bar: goal:MakeReservation continuation:MakeReservation. For example:

  • Pick a different habitat pod

  • Change that to 2 astronauts

  • Select a different date

We train these as "Continuation of" MakeReservation for cases where the users are at the Confirmation screen to review their order and decide that they want to make some changes. The goal remains MakeReservation, and any Values are annotated as such (ex: number of astronauts, pod name). This time, the flagged signal route is to ChangeOrder. This is to re-route the request to update the Order with the newly provided information. For generic requests that do not contain a new input Value (ex: Change the number of astronauts), we add an extra flagged signal route to the action for that request (ex: GetNumberOfAstronauts).

Prompting Flows

Confirmation Prompt

Confirmation Prompt

You can see all trained utterances and plans by entering this query in the training tab search bar: prompt:Confirmation. For example:

  • Yes

  • Do it

When the user is done reviewing their Order at the Confirmation Prompt, they can use these utterances to move forward and proceed with the reservation. This is an "At prompt for" Confirmation with goal Confirmation. The Confirmation itself is annotated with a boolean Value "true" or "false".

SpaceResort Selection Prompt

SpaceResort Selection Prompt

You can see all trained utterances and plans by entering this query in the training tab search bar: prompt:SpaceResort. For example:

  • The one with a refueling station

  • The Mercurial

  • The one on Venus

The booking flow only allows a single SpaceResort at a time, so when there are multiple candidates, the user will be presented with a SpaceResort Selection Prompt. For prompt training, the goal must always match the prompt context, so we train these as "At prompt for" SpaceResort with goal SpaceResort. There are many ways the user can answer, so we annotate any provided Value (space resort name, planet, search criteria) and add a special flagged signal route to the SelectResort action. This action will take the hotels currently in context and attempt to filter them based on the newly provided inputs. For example:

  • User: Book a hotel near Jupiter

  • Bixby: Here are some hotels around Jupiter. Which one would you like?

  • User: The one with a refueling station.

  • Bixby: There are 3 hotels around Jupiter with a refueling station. Which one? (Where the 3 options are a subset of the previous options, not a new search)

  • User: The Ganymede Moon Motel

  • Bixby: (Proceeds with the booking flow)

Other Selection Prompts (NumberOfAstronauts, HabitatPod, DateInterval...)

HabitatPod Selection Prompt

You can see all trained utterances and plans by entering this query in the training tab search bar: has:prompt -goal:SpaceResort -goal:Confirmation. For example:

  • 3 astronauts

  • The Honey Moon Suite

  • Next weekend

For prompt training, the goal must always match the prompt context, so we train these as "At prompt for" <Concept> and goal <Concept>. Then we annotate the Value in the utterance for that <Concept>. For example, "At prompt for" HabitatPod has goal HabitatPod and the PodName is annotated as a value.

Additional Resources

Your Source for Everything Bixby

Guides & Best Practices

Bixby Videos

Bixby Podcast

Bixby on Social Media

Need Support?

  • Have a feature request? Please suggest it in our Support Community to help us prioritize.
  • Have a technical question? Ask on Stack Overflow with tag “bixby”
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].