Hands-on Primer on Rasa

Hands-on Primer on Rasa - Part 3

Previously in Part-2 of this series, we configured, trained, and deployed a simple helper chatbot instance using the Rasa chatbot framework.

For the simple helper chatbot, the conversational universe was controlled via Rasa intents and stories. In other words, to expand the conversational domain, one would add more intents and stories, followed by the model training and then the chatbot deployment.

What if one wanted to expand the chatbot support to respond to more linux commands ??? The approach using intents and stories would be tedious and time consuming. Is there a better approach for dynamic expansion ???

This is where the Rasa custom actions (via the Rasa SDK) comes into play. With Rasa custom actions, one can enable Python code to be executed on behalf of the users (based on a conversation), to perform tasks beyond the pre-defined intents and stories.

In third part of this primer, we will enhance the helper chatbot using custom actions, to enable dynamic addition of more linux command responses.

Note that for this demonstration, we will store the linux command responses in a pipe-delimited text file for simplicity instead of a database.

First, the file config.yml is used to define the behavior of Rasa NLU and Rasa Core sub-systems.

We will keep the contents of the file unchanged as shown below:

config.yml

# Configuration for Rasa NLU.
# https://rasa.com/docs/rasa-pro/nlu-based-assistants/components/
language: en

pipeline:
# # See https://rasa.com/docs/rasa-pro/nlu-based-assistants/model-configuration for more information.
    - name: WhitespaceTokenizer
    - name: RegexFeaturizer
    - name: LexicalSyntacticFeaturizer
    - name: CountVectorsFeaturizer
    - name: CountVectorsFeaturizer
      analyzer: char_wb
      min_ngram: 1
      max_ngram: 4
    - name: DIETClassifier
      epochs: 100
      constrain_similarities: true
    - name: EntitySynonymMapper
    - name: ResponseSelector
      epochs: 100
      constrain_similarities: true
    - name: FallbackClassifier
      threshold: 0.3
      ambiguity_threshold: 0.1

# Configuration for Rasa Core.
# https://rasa.com/docs/rasa-pro/concepts/policies/policy-overview/

policies:
# # See https://rasa.com/docs/rasa-pro/concepts/policies/policy-overview for more information.
    - name: MemoizationPolicy
    - name: RulePolicy
    - name: UnexpecTEDIntentPolicy
      max_history: 5
      epochs: 100
    - name: TEDPolicy
      max_history: 5
      epochs: 100
      constrain_similarities: true

Note that the DIETClassifier not only classifies the intent, but also extracts entities.

Next, the file domain.yml is used to define the universe of the Rasa chatbot, such as, the intents and entities identified by the Rasa NLU sub-system, the slots to be remembered, the actions to be executed on behalf of the users, and the responses to send to the users based on the intent.

We will change the contents of the file as shown below:

domain.yml

version: "3.1"

intents:
  - greet
  - goodbye
  - linux_command_help

entities:
  - command_ask

slots:
  command_ask:
    type: text
    influence_conversation: false
    mappings:
      - type: from_entity
        entity: command_ask

actions:
  - action_linux_commands

responses:
  utter_greet:
  - text: "Howdy!"

  utter_ready_to_help:
  - text: "What linux command would you need help with?"

  utter_command_ask_unclear:
  - text: "Not sure what linux command you need help with?"

  utter_goodbye:
  - text: "Bye"

session_config:
  session_expiration_time: 60
  carry_over_slots_to_new_session: true

Moving along, the file data/nlu.yml is used to define the training samples of all the user intents, along with examples of the utterances corresponding to the intents, for the Rasa chatbot to be trained on.

We will change the contents of the file as shown below:

data/nlu.yml

version: "3.1"

nlu:

- intent: greet
  examples: |
    - hey
    - hello
    - hi
    - hello there
    - hey there

- intent: goodbye
  examples: |
    - good bye
    - bye
    - goodbye
    - bye bye
    - see you later

- intent: linux_command_help
  examples: |
    - check [host reachable](command_ask)
    - display [network interfaces](command_ask)
    - find [directory size](command_ask)
    - get [host info](command_ask)
    - how to [search keyword](command_ask)
    - list [mounted devices](command_ask)
    - show [network interfaces](command_ask)

Notice that the number intents have been reduced and replaced by a new one with entity references. The syntax [value](name) enables Rasa NLU to extract the 'value' and assign it to the entity slot 'name' from the users' messages. In our example, the entity is referred to as command_ask.

Next stop, the file data/rules.yml is used to define the training data for all conversations that always follow a pre-defined set of steps, which the Rasa chatbot can be trained on.

We will change the contents of the file as shown below:

data/rules.yml

version: "3.1"

rules:

- rule: Say goodbye anytime the user says goodbye
  steps:
  - intent: goodbye
  - action: utter_goodbye

- rule: Ask about a linux command
  steps:
  - intent: linux_command_help
  - action: action_linux_commands

Notice that the we have added an additional rule which will invoke the custom action at runtime.

Moving on, the file data/stories.yml is used to define the training data for all the dialog scenarios, which represent the users intent to the appropriate Rasa chatbot response.

We will change the contents of the file as shown below:

data/stories.yml

version: "3.1"

stories:

- story: greet path
  steps:
  - intent: greet
  - action: utter_greet
  - action: utter_ready_to_help

Notice that the we have removed the pre-defined story paths and simplified the flow.

Finally, the file endpoints.yml is used to define all the network service(s) the Rasa chatbot can connect to. We will run our custom action as a separate webhook service.

We will change the contents of the file as shown below:

endpoints.yml

# This file contains the different endpoints your bot can use.

# Server where the models are pulled from.
# https://rasa.com/docs/rasa-pro/production/model-storage#fetching-models-from-a-server

# Server which runs your custom actions.
# https://rasa.com/docs/rasa-pro/concepts/custom-actions

action_endpoint:
  url: "http://192.168.1.25:5055/webhook"

Now, it is time to train our enhanced linux commands' helper chatbot model.

Before we kick the training process, execute the following command in the terminal window:

$ ls -l $HOME/.rasa/models

The following should be the typical output:

Output.1

total 47672
-rw-r--r-- 1 polarsparc polarsparc 24992178 Nov  2 14:17 20241102-181642-avocado-bright.tar.gz
-rw-r--r-- 1 polarsparc polarsparc 23820630 Nov  9 19:13 20241110-001312-approximate-beech.tar.gz

To train the enhanced linux commands' helper chatbot, execute the following command in the terminal window:

$ docker run --rm --name rasa-pro -u $(id -u $USER):$(id -g $USER) -v $HOME/.rasa/:/app -e RASA_PRO_LICENSE=${RASA_PRO_LICENSE} rasa/rasa-pro:3.10.8 train --verbose

The following should be the typical trimmed output:

Output.2

[ ... SNIP ... ]
2024-11-11 19:21:12 INFO     rasa.cli.train  - {"event_info": "Started validating domain and training data...", "event": "cli.train.run_training", "level": "info"}
[ ... SNIP ... ]
2024-11-11 19:21:17 INFO     rasa.validator  - {"event_info": "Validating intents...", "event": "validator.verify_intents_in_stories.start", "level": "info"}
2024-11-11 19:21:17 INFO     rasa.validator  - {"event_info": "Validating uniqueness of intents and stories...", "event": "validator.verify_example_repetition_in_intents.start", "level": "info"}
2024-11-11 19:21:17 INFO     rasa.validator  - {"event_info": "Story structure validation...", "event": "validator.verify_story_structure.start", "level": "info"}
Processed story blocks: 100%|||||||||||| 1/1 [00:00<00:00, 3141.80it/s, # trackers=1]
2024-11-11 19:21:17 INFO     rasa.core.training.story_conflict  - Considering all preceding turns for conflict analysis.
2024-11-11 19:21:17 INFO     rasa.validator  - {"event_info": "No story structure conflicts found.", "event": "validator.verify_story_structure.no_conflicts", "level": "info"}
2024-11-11 19:21:17 INFO     rasa.validator  - {"event": "validation.flows.started", "level": "info"}
2024-11-11 19:21:17 WARNING  rasa.validator  - {"event_info": "No flows were found in the data files. Will not proceed with flow validation.", "event": "validator.verify_flows", "level": "warning"}
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Starting to train component 'RegexFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Finished training component 'RegexFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Starting to train component 'LexicalSyntacticFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Finished training component 'LexicalSyntacticFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Starting to train component 'CountVectorsFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.nlu.featurizers.sparse_featurizer.count_vectors_featurizer  - 29 vocabulary items were created for text attribute.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Finished training component 'CountVectorsFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Starting to train component 'CountVectorsFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.nlu.featurizers.sparse_featurizer.count_vectors_featurizer  - 369 vocabulary items were created for text attribute.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Finished training component 'CountVectorsFeaturizer'.
2024-11-11 19:21:18 INFO     rasa.engine.training.hooks  - Starting to train component 'DIETClassifier'.
Epochs: 100%|||||||||||| 100/100 [00:25<00:00,  3.93it/s, t_loss=1.04, i_acc=1, e_f1=1]       
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Finished training component 'DIETClassifier'.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Starting to train component 'EntitySynonymMapper'.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Finished training component 'EntitySynonymMapper'.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Starting to train component 'ResponseSelector'.
2024-11-11 19:21:44 INFO     rasa.nlu.selectors.response_selector  - Retrieval intent parameter was left to its default value. This response selector will be trained on training examples combining all retrieval intents.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Finished training component 'ResponseSelector'.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Restored component 'MemoizationPolicy' from cache.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Restored component 'RulePolicy' from cache.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Restored component 'TEDPolicy' from cache.
2024-11-11 19:21:44 INFO     rasa.engine.training.hooks  - Restored component 'UnexpecTEDIntentPolicy' from cache.
2024-11-11 19:21:47 INFO     rasa.model_training  - {"event_info": "Your Rasa model is trained and saved at 'models/20241111-192118-oily-sedan.tar.gz'.", "event": "model_training.train.finished_training", "level": "info"}

The training process has completed and one more time execute the following command in the terminal window:

$ ls -l $HOME/.rasa/models

The following should be the typical output:

Output.3

total 71072
-rw-r--r-- 1 polarsparc polarsparc 24992178 Nov  2 14:17 20241102-181642-avocado-bright.tar.gz
-rw-r--r-- 1 polarsparc polarsparc 23820630 Nov  9 19:13 20241110-001312-approximate-beech.tar.gz
-rw-r--r-- 1 polarsparc polarsparc 23960583 Nov 11 14:21 20241111-192118-oily-sedan.tar.gz

Notice that a new chatbot model 20241111-192118-oily-sedan.tar.gz has been created.

Now, we are ready to unveil the most interesting part of this primer - the Python code for the Rasa custom action.

The following is the Python code for the custom action:

actions/actions.py

#
# @Author: Bhaskar S
# @Blog:   https://www.polarsparc.com
# @Date:   10 Nov 2024
#

import csv
import logging

from typing import Any, Dict, List, Text

from rasa_sdk import Action, Tracker
from rasa_sdk.executor import CollectingDispatcher

logging.basicConfig(format='%(asctime)s - %(message)s', level=logging.INFO)


class ActionLinuxCommandsHelper(Action):
    def __init__(self):
        self.commands = {}
        super(ActionLinuxCommandsHelper, self).__init__()
        # Change to './commands.txt' for Testing
        with open('/app/actions/commands.txt', 'r') as csvfile:
            reader = csv.reader(csvfile, delimiter='|')
            for row in reader:
                self.commands[row[0].lower()] = row[1]
        logging.info(f'Action linux commands loaded with keys -> {self.commands.keys()}')

    def name(self) -> Text:
        return 'action_linux_commands'

    def run(self,
            dispatcher: CollectingDispatcher,
            tracker: Tracker,
            domain: Dict[Text, Any]) -> List[Dict[Text, Any]]:
        command_ask = next(tracker.get_latest_entity_values('command_ask'), None)

        logging.info(f'User command ask -> {command_ask}')

        if not command_ask:
            dispatcher.utter_message(response='utter_command_ask_unclear')
        else:
            command_text_msg = self.commands.get(command_ask.lower(), None)

            logging.info(f'Identified command text -> {command_text_msg}')

            if not command_text_msg:
                dispatcher.utter_message(response='utter_command_ask_unclear')
            else:
                dispatcher.utter_message(text=command_text_msg)
        return []


# Only for Testing
if __name__ == '__main__':
    my_tracker = Tracker('test', {'command_ask': 'directory size'}, None,
                          None, None, None, None, None, None)
    my_dispatcher = CollectingDispatcher()
    commands_helper = ActionLinuxCommandsHelper()
    commands_helper.run(my_dispatcher, my_tracker, None)

Every custom action is implemented as a Python class, which must inherit the base class rasa_sdk.Action. The custom action class must implement the following two methods:

name(self) :: the value returned from this method MUST match the name of the custom action defined in the file domain.yml
run(self, dispatcher, tracker, domain) :: is invoked each time the intent of the user input message matches the intent of this custom action. The tracker object provides access to the entity slot(s) and the dispatcher object is used to send a response back to the user

Notice that we use the contents of the pipe-delimited file commands.txt to initialize an in-memory Python dictionary.

The following are the contents of the file commands.txt:

actions/commands.txt

directory size|Use the command 'du -s -m <directory>' to find the size of the directory and its contents
mounted devices|Use the command 'df -h' to find the information about all the mounted devices
search keyword|Use the command 'grep -i keyword <file>' to search for all the lines with the specified keyword
find file|Use the command 'find <directory> -type f -name <file>' to search for the specified file in the given directory
host reachable|Use the command 'ping -c 3 <host>' to check if the host is reachable
host info|Use the command 'uname -a' to get information about the host or system
network interfaces|Use the command 'netstat -i' to display information about all the network interfaces in the system

To launch the custom action as a webhook service, execute the following command in the terminal window:

$ docker run -it --rm --name rasa-pro-action --network="host" -u $(id -u $USER):$(id -g $USER) -v $HOME/.rasa:/app -e RASA_PRO_LICENSE=${RASA_PRO_LICENSE} -e SANIC_HOST=192.168.1.25 rasa/rasa-pro:3.10.8 run actions

The following should be the typical output:

Output.4

2024-11-11 19:23:30 INFO     rasa.tracing.config  - No endpoint for tracing type available in endpoints.yml,tracing will not be configured.
2024-11-11 19:23:31 INFO     root  - Action linux commands loaded with keys -> dict_keys(['directory size', 'mounted devices', 'search keyword', 'find file', 'host reachable', 'host info', 'network interfaces'])
2024-11-11 19:23:31 INFO     rasa_sdk.executor  - Registered function for 'action_linux_commands'.
2024-11-11 19:23:31 INFO     rasa_sdk.endpoint  - Starting action endpoint server...
2024-11-11 19:23:31 INFO     rasa_sdk.endpoint  - Starting plugins...
2024-11-11 19:23:31 INFO     rasa_sdk.endpoint  - Action endpoint is up and running on http://192.168.1.25:5055
2024-11-11 19:23:31 INFO     rasa_sdk.tracing.config  - No endpoint for tracing type available in endpoints.yml,tracing will not be configured.
2024-11-11 19:23:31 INFO     sanic.server  - Starting worker [1]

To test the enhanced linux commands' helper chatbot, execute the following command in the terminal window:

$ docker run -it --rm --name rasa-pro -u $(id -u $USER):$(id -g $USER) -v $HOME/.rasa/:/app -e RASA_PRO_LICENSE=${RASA_PRO_LICENSE} rasa/rasa-pro:3.10.8 shell

The following should be the typical trimmed output:

Output.5

[ ... SNIP ... ]
2024-11-11 18:35:34 INFO     root  - Connecting to channel 'cmdline' which was specified by the '--connector' argument. Any other channels will be ignored. To connect to all given channels, omit the '--connector' argument.
2024-11-11 18:35:34 INFO     root  - Starting Rasa server on http://0.0.0.0:5005
2024-11-11 18:35:35 INFO     rasa.core.processor  - Loading model models/20241111-183402-bent-goal.tar.gz...
[ ... SNIP ... ]
2024-11-11 18:36:08 INFO     root  - Rasa server is up and running.
Bot loaded. Type a message and press enter (use '/stop' to exit): 
Your input ->

Notice that the enhanced linux commands' helper chatbot is waiting for user input on the prompt Your input ->.

Type hello at the chatbot prompt.

The following should be the typical output:

Output.6

Howdy!
What linux command would you need help with?

Next, type list mounted devices at the chatbot prompt.

The following should be the typical output:

Output.7

Use the command 'df -h' to find the information about all the mounted devices

Next, type show network interfaces at the chatbot prompt.

The following should be the typical output:

Output.8

Use the command 'netstat -i' to display information about all the network interfaces in the system

Next, type show running processes at the chatbot prompt.

The following should be the typical output:

Output.9

Not sure what linux command you need help with?

There is no matching entry in the commands.txt file and hence this response.

Stop the custom action webhook service.

Modify the contents of the file commands.txt by adding a new entry as follows:

actions/commands.txt

directory size|Use the command 'du -s -m <directory>' to find the size of the directory and its contents
mounted devices|Use the command 'df -h' to find the information about all the mounted devices
search keyword|Use the command 'grep -i keyword <file>' to search for all the lines with the specified keyword
find file|Use the command 'find <directory> -type f -name <file>' to search for the specified file in the given directory
host reachable|Use the command 'ping -c 3 <host>' to check if the host is reachable
host info|Use the command 'uname -a' to get information about the host or system
network interfaces|Use the command 'netstat -i' to display information about all the network interfaces in the system
running processes|Use the command 'ps -fu' to display all the running processes

Now, restart the custom action webhook service.

Next, type show running processes at the chatbot prompt.

The following should be the typical output:

Output.10

Use the command 'ps -fu' to display all the running processes

Finally, type /stop at the chatbot prompt and the chatbot will exit.

YIPEE - we have successfully tested our enhanced linux commands' helper chatbot !!!

Given that we have covered the creation and demonstration of an enhanced linux commands' helper chatbot using custom actions, we conclude Part-3 of this series !!!