This repo contains sample code for building task-oriented conversational bots using Project Conversation Learner, and demonstrating how to use the @conversationlearner/sdk
npm module.
Project Conversation Learner enables you to build and teach task-oriented conversational interfaces that learn from example interactions, combining code, logical constraints and machine learning. Certain absolute parts of your application such as checking if the user is logged in or making an API request to check store inventory can still be coded; however, other changes in state and action selection can be learned from example dialogs given by the domain expert or developer.
Cognitive Service Labs provides developers with an early look at emerging Cognitive Services technologies. Early adopters who do not need market-ready technology can discover, try and provide feedback on new Cognitive Services technologies before they are generally available. Labs are not Azure services. Project Conversation Learner’s APIs and schema are subject to change that are not backwards compatible.
This README explains how to get started; full documentation is available here.
An invitation is required to access Project Conversation Learner.
Project Conversation Learner consists of an SDK you add to your bot, and a cloud service which the SDK accesses for machine learning. At present, access to the Project Conversation Leaner cloud service requires an invitation. If you haven't been invited already, request an invitation. If you have not received an invitation, you will be unable to access the cloud API.
-
An invitation to access the Project Conversation Learner cloud service -- request an invitation.
-
Node 8.5.0 or higher and npm 5.3.0 or higher. Install from https://nodejs.org/en/
-
LUIS authoring key:
-
Log into http://www.luis.ai
-
Click on your name in the upper right, then on "settings"
-
Authoring key is shown on the resulting page
(Your LUIS authoring key serves 2 roles. First, it will serve as your Conversation Learner authoring key. Second, Conversation Learner uses LUIS for entity extraction; the LUIS authoring key is used to create LUIS models on your behalf)
-
-
Google Chrome web browser. Install from https://www.google.com/chrome/index.html.
-
git. Install from https://git-scm.com/downloads.
-
VSCode. Install from https://code.visualstudio.com/. Note this is recommended, not required.
-
Install and build:
git clone https://github.com/Microsoft/ConversationLearner-Samples my-bot-01 cd my-bot-01 npm install npm run build
Note: during
npm install
, you can ignore this error if it occurs:gyp ERR! stack Error: Can't find Python executable
-
Configure:
Create a file called
.env
in the directorymy-bot-01
. The contents of the file should be:NODE_ENV=development LUIS_AUTHORING_KEY=<your LUIS authoring key>
-
Start bot:
npm start
This runs the generic empty bot in
my-bot-01/src/app.ts
. -
Run Conversation Learner UI:
[open second command prompt window] cd my-bot-01 npm run ui
-
Open browser to http://localhost:5050
You're now using Conversation Learner and can create and teach a Conversation Learner model.
The instructions above started the generic empty bot. To run a tutorial or demo bot instead:
-
If you have the Conversation Learner web UI open, return to the list of models at http://localhost:5050/home.
-
If another bot is running (like
npm start
ornpm run demo-pizza
), stop it. You do not need to stop the UI process, or close the web browser. -
Run a demo bot from the command line (step 2 above). Demos include:
npm run tutorial-general
npm run tutorial-entity-detection
npm run tutorial-session-callbacks
npm run tutorial-api-calls
npm run demo-password
npm run demo-pizza
npm run demo-storage
npm run demo-vrapp
-
If you're not already, switch to the Conversation Learner web UI in Chrome by loading http://localhost:5050/home.
-
Click on "Import tutorials" (only needs to be done once). This will take about a minute and will copy the Conversation Learner models for all the tutorials into your Conversation Learner account.
-
Click on the demo model in the Conversation Learner UI that corresponds to the demo you started.
Source files for the demos are in my-bot-01/src/demos
-
If you have the Conversation Learner web UI open, return to the list of models at http://localhost:5050/home.
-
If a bot is running (like
npm run demo-pizza
), stop it. You do not need to stop the UI process, or close the web browser. -
If desired, edit code in
my-bot-01/src/app.ts
. -
Rebuild and re-start bot:
npm run build npm start
-
If you're not already, switch to the Conversation Learner web UI in Chrome by loading http://localhost:5050/home.
-
Create a new Conversation Learner application in the UI, and start teaching.
-
To make code changes in
my-bot-01/src/app.ts
, repeat the steps above, starting from step 2.
In VSCode, there are run configurations for each demo, and for the "Empty bot" in my-bot-01/src/app.ts
. Open the my-bot-01
folder in VSCode.
There is a template .env.example
file shows what environment variables you may set to configure the samples.
You can adjust these ports to avoid conflicts between other services running on your machine by adding a .env
file to root of project:
cp .env.example .env
This uses the standard configuration, which lets you run your bot locally, and start using Conversation Learner. (Later on, to deploy your bot to the Bot Framework, some edits to this file will be needed.)
Publish your Conversation Learner bot similar to the way you would publish any other bot. At a high level, you upload your code to a hosted website, set the appropriate configuration values, and then register the bot with various channels. Below we cover steps for deploying your bot to Azure and create a Bot Framework Channel Registration for it.
Your Conversation Learner bot is a Node.js app and deploying it to Azure is not different than deploying a Node.js application or Node.js Bot Framework bot. You can read more about deploying a Node.js web application to Azure here.
Below we cover step-by-step instructions to deploy your bot to Azure in two different ways: 1) ZipDeploy using KuduEngine 2) Deploy from local git repository
ZipDeploy using KuduEngine
- Login to the Azure portal
- Create a new "Web App" using Azure app services
- Give your app/bot a name
- Select/Create a resource group for your app
- Use "Windows" as Web application's OS
- Once the step above is successful, you will end up with an empty web application deployed for you under Azure app services. Select the app service and navigate to its dashboard.
- Set the following Application Settings for your app service
-
Click on "Application Settings" in the navigation pane
-
Scroll down to the "Application Settings" section
-
Add these settings:
APP SETTING NAME VALUE CONVERSATION_LEARNER_MODEL_ID Model Id GUID, obtained from the Conversation Learner UI under the "settings" for the app> LUIS_AUTHORING_KEY LUIS authoring key for this app SCM_DO_BUILD_DURING_DEPLOYMENT true -
Click on "Save" near the top of the page
-
- From the app service dashboard's navigation pane select the
advanced tools
and navigate to Kudu website for your app service. The url for Kudu tools of your web application ishttps://<your_app_name>.scm.azurewebsites.net
- Zip local folder of your Conversation Learner bot (Note that zip file should have all the artifacts in its root and not in a subdirectory). You don't need to include the following folders/files in the zip file: 1) Any folder/file starting with
.
, e.g. .git, .github, 2) node_modules folder. (SCM_DO_BUILD_DURING_DEPLOYMENT
application settings will configure Kudu deployment engine to runnpm install
as part of deployment and install all the necessary node dependencies based on package.json. You can read more about deploy from a zip file here.). Also make sure that you have the engine version configured in your package.json to node.js 8.8 or higher. If not make sure to set theWEBSITE_NODE_DEFAULT_VERSION
to8.8
in application settings. - Navigate to
https://<your_app_name>.scm.azurewebsites.net/ZipDeploy
and drop the zip file from previous step to the file explorer area in the website - After a successful deployment, you should be able to send messages to your deployed bot messaging endpoint:
https://<your_app_name>.azurewebsites.net/api/messages
using bot framework emulator. You can read more about using Bot Framework emulator here.
Follow the steps 1 to 4 above to create a web application
on Azure and then follow the steps bellow:
- Go to
Deployment Center
of your web application and configure deployment withLocal Git
as the source control andApp Service Kudu build server
as the build provider. Once the deployment fromLocal Git
is configured successfully, you will have a git repository with the following endpoint for your app service:https://<your_app_name>.scm.azurewebsites.net/<your_app_name>.git
. Make sure to configure the deployment credentials for your web application, so you can push to it from your local repository in the next steps - add the azure remote to your local git repository:
git remote add azure https://<your_app_name>.scm.azurewebsites.net/<your_app_name>.git
- Push your local repository to azure remote:
git push azure <your_local_branch>:master
(If you are using typescript for development, make sure to commit the build artifacts, i.e. *.js files, to your local repository.) - After a successful deployment, you should be able to send messages to your deployed bot endpoint:
https://<your_app_name>.azurewebsites.net/api/messages
using bot framework emulator
After successful deployment of your Conversation Learner bot as an Azure app service, you can follow these steps to connect it to a Bot Framework channel:
-
Select
Bot Services
from azure portal services and add a newBot Service
-
Select
Bot Channel Registration
from the availableBot Service
options -
Fill out the necessary fields for your channel registration, e.g.
bot name
. For the messaging endpoint provide the url from the successful bot deployment you have done by following the steps from previous section. The messaging endpoint will be in the following format:https://<your_app_name>.azurewebsites.net/api/messages
. Also you need to create a Microsoft App Id and Microsoft App Password for your bot. Preserve the Microsoft app id and password that you are creating for your channel registration. -
Set the following
Application Settings
for your web application:APP SETTING NAME VALUE MicrosoftAppId Application Id GUID from previous step MicrosoftAppPassword Application password from previous step -
You should be able to test your bot using Bot Framework web chat
Note: Ensure that these variables are set when deploying your bot:
Environment variable | Setting |
---|---|
CONVERSATION_LEARNER_MODEL_ID | Model Id GUID, obtained from the Conversation Learner UI under the "settings" for the app |
LUIS_AUTHORING_KEY | LUIS authoring key for this app. Obtained from https://www.luis.ai |
MicrosoftAppId | Microsoft Application Id |
MicrosoftAppPassword | Microsoft Application Password |
Once the bot is deployed and running you can connect different channels to it such as Facebook, Teams, Skype etc using an Azure Bot Channel Registration. For documentation on that process see: https://docs.microsoft.com/en-us/bot-framework/bot-service-quickstart-registration
- Tag questions on Stack Overflow with "microsoft cognitive"
- Request a feature on our User Voice page
- Open an issue on our github repo
When submitting PR's please target the 'develop' branch unless it is a hotfix for 'master'
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.