Match latest getting started flow (discussion)

.travis.yml

@@ -2,17 +2,17 @@ language: python
 sudo: required
 services:
 - docker
+- xvfb
 python:
   - '3.6'
 before_install:
 - python -m pip install -U pip
 - pip install -e . --no-use-pep517
 - export DISPLAY=0.0
-- make train-memo
+- rasa train core --config augmentedmemo-only.yml --augmentation 0 --quiet
 - "export DISPLAY=:99.0"
-- "sh -e /etc/init.d/xvfb start"
 - sleep 3 # give xvfb some time to start
-- make evaluate-core
+- rasa test core --fail-on-prediction-errors --quiet
 - openssl aes-256-cbc -K $encrypted_8857e34a5e52_key -iv $encrypted_8857e34a5e52_iv
   -in travis_gcloud_auth.json.enc -out travis_gcloud_auth.json -d
 - docker login -u _json_key -p "$(cat travis_gcloud_auth.json)" https://gcr.io

README.md

@@ -4,10 +4,11 @@
 ## :surfer: Introduction
 The purpose of this repo is to showcase a contextual AI assistant built with the open source Rasa framework.
 
-Sara is an alpha version and lives in our docs (temporarily she's unavailable on the docs due to maintenance), helping developers getting started with our open source tools. It supports the following user goals:
+Sara is an alpha version and lives in our docs (temporarily she's unavailable on the docs due to maintenance), 
+helping developers getting started with our open source tools. It supports the following user goals:
 
 - Understanding the Rasa framework
-- Installing the Rasa framework
+- Getting started with Rasa
 - Answering some FAQs around Rasa
 - Subscribing to the Rasa newsletter
 - Requesting a call with Rasa's sales team
@@ -16,32 +17,43 @@ Sara is an alpha version and lives in our docs (temporarily she's unavailable on
 You can talk to Sara [here](https://rasa.com/docs/get_started_step1/) (temporarily unavailable) and find planned enhancements for Sara in the
 [Project Board](https://github.com/RasaHQ/rasa-demo/projects/1)
 
-## 🤖 How to install and run Sara
+## 👷‍ Installation
 
 To install Sara, please clone the repo and run:
 
 ```
 cd rasa-demo
+pip install -r requirements.txt
 pip install -e .
 ```
 This will install the bot and all of its requirements.
-Note that it was written in Python 3 so might not work with PY2.
+Note that this bot should be used with python 3.6 or 3.7.
 
-To train the model: `make train` (this will take a significant amount of memory to train,
+## 🤖 To run Sara:
+
+Use `rasa train` to train a model (this will take a significant amount of memory to train,
 if you want to train it faster, try the training command with
-`--augmentation 0`)
+`--augmentation 0`).
 
-To run Sara with both these models:
-```
-docker run -p 8000:8000 rasa/duckling
-make run-cmdline
+Then, to run, first set up your action server in one terminal window:
+```bash
+rasa run actions --actions demo.actions
 ```
 
 There are some custom actions that require connections to external services,
-specifically `ActionSubscribeNewsletter` and `ActionStoreSalesInfo`. For these
+specifically `SubscribeNewsletterForm` and `SalesForm`. For these
 to run you would need to have your own MailChimp newsletter and a Google sheet
 to connect to.
 
+In another window, run the bot:
+```bash
+docker run -p 8000:8000 rasa/duckling
+rasa shell --debug
+```
+
+Note that `--debug` mode will produce a lot of output meant to help you understand how the bot is working 
+under the hood. To simply talk to the bot, you can remove this flag.
+
 If you would like to run Sara on your website, follow the instructions
 [here](https://github.com/mrbot-ai/rasa-webchat) to place the chat widget on
 your website.
@@ -50,15 +62,15 @@ your website.
 
 `data/core/` - contains stories 
 
-`data/nlu` - contains example NLU training data
+`data/nlu` - contains NLU training data
 
 `demo` - contains custom action/api code
 
-`domain.yml` - the domain file 
+`domain.yml` - the domain file, including bot response templates
 
-`config.yml` - the Rasa config file
+`config.yml` - training configurations for the NLU pipeline and policy ensemble
 
-### Code Style
+## ⚫️ Code Style
 
 To ensure a standardized code style we use the formatter [black](https://github.com/ambv/black).
 

config.yml

@@ -5,8 +5,13 @@ pipeline:
 - name: CountVectorsFeaturizer
   OOV_token: oov
   token_pattern: (?u)\b\w+\b
+- name: CountVectorsFeaturizer
+  analyzer: "char_wb"
+  min_ngram: 1
+  max_ngram: 4
 - name: EmbeddingIntentClassifier
   epochs: 50
+  loss_type: margin
 - name: DucklingHTTPExtractor
   url: http://localhost:8000
   dimensions: