Skip to main content
Skip table of contents

Controller Bot Training

Context

In this document we will explain the overall controller training and also explain how we add new training for controller.

Controller Training

In controller there are different training for controller which already configure in it and we also add new training file.First we discuss the the already configure training's and after that we see that how we add new training to the controller. Before explaining the list of training event / action  we explain the common things which is use in all the training.

All the training file is basically is a python file and all are containing a method called run which accept some parameter which are listed below.

ParameterDescription
self the self variable is a commonly used convention for referring to the instance of a class within its own methods
conversationconversation parameter contain an object of conversation which contain all info about conversation
slotsslots parameter contain attribute in key value pair for a conversation / event .we also add slots in a conversation.
dispatcherdispatcher is basically a respond back action from a controller
metadatametadata contain metadata of a conversation its not be used in training

Here is the list of all the configure training events:

EventDescription
CONVERSATION_STARTEDIn this training action we first get the channel session from slot and than we get bot id from that channel session and set the role is PRIMARY. After that we respond back an action with the help of dispatcher name ASSIGN_BOT with bot id and role.
AGENT_SUBSCRIBEIn this training action we get conversation id from conversation and agent id from slot agentParticipant->participant→id. If there is no channel session than we respond back an action with the help of dispatcher name REVOKE_REQUEST with reasonCode CANCELLED and agent_id. After that we return by set the slot of agent_state to not_requested. If there is a channel session than we get bot participant from conversation and if there is bot participant and its role is PRIMARY than we get bot id and respond bank an action by dispatcher name CHANGE_PARTICIPANT_ROLE with role is ASSISTANT and bot id latter we set the slot agent_state to subscribed.
AGENT_UNSUBSCRIBEIn this training action we first get conversation id from conversation after that we get cc_user list and channel session list from conversation. If there is no cc_user mean there is no agent all the agent are left the conversation we get the agent_state from slot and direction from agent_state if the agent_state is "requested" and direction is "DIRECT_CONFERENCE" than we respond back an action by dispatcher name CANCEL_RESOURCE with reasonCode CANCELLED. After that we check that if there is no channel session main all the customer are left that we respond back an action by dispatcher name END_CONVERSATION and return with type reset. Otherwise if customer is still present than we check if there is a bot participant than change his role PRIMARY with the help of dispatcher action name CHANGE_PARTICIPANT_ROLE with role. After that we get schedule timer as events from slot and conversation . If agent_state is requested and direction is DIRECT_TRANSFER than return that events. After that we get routing mode from conversation channel session and reason code from slot agentSubUnSubReason. If routing_mode is PUSH and reason code is FORCE_LOGOUT than dispatcher an action name FIND_AGENT and append event with set slot agent_state to request and return that events .Otherwise append event with set slot agent_state to not_requested and return that events. 
CHANNEL_SESSION_STARTEDIn this training we first get conversation id from conversation after that we get channel session from slot and channel sessions list from conversation. After that we get channel_session_sla_map from slot key and set channel session['id'] in channel_session_sla_map to False. After that we schedule timer for all sessions as a event .After that we get latestIntent from channel session if there is no channel session than dispatcher a text "Hey! this is sparrow from Expert Flow. How may i help you today?" and return all events
CHANNEL_SESSION_EXPIREDIn this training we first get conversation id from conversation ,channel session from slot ,channel session list from conversation and also get cc_user list from conversation. If there is no channel session list mean customer is left than we get agent state from slot and if agent state is requested that we respond back with a dispatcher action name CANCEL_RESOURCE with reasonCode CANCELLED and append events array with slot agent_state. After that if there is no cc_user mean all agent are left that we call a dispatcher action name END_CONVERSATION and return with type reset. ......................................
TASK_STATE_CHANGEDIn this training we first check is there is a conversation or not if there is no conversation than do nothing else if there is a conversation then get taskDto from slots and get task_state from taskDto if task_state name is CLOSED and reasonCode is FORCE_CLOSED then get task_type from taskDto if task type direction is DIRECT_CONFERENCE and mode QUEUE then dispatcher a text "'Your chat has ended due to technical issues'.'Please come back later at a convenient time'" and set slot agent_state and return.
end_chatIn this training we first get conversation id and channel session from conversation if there is no channel session then just console info log that there is no channel session unable to post channel session expired. Otherwise dispatcher an action with name CHANNEL_SESSION_EXPIRED with channel session.
defaultIn default training we just log info that default action called.

How to add new training:

  1. Add training event into the action_mapper.py class with his mapper class.
  2. Create training event class which run method with a parameter as we mentioned above.
  3. Write the functionality into that run method.
  4. Save the training file and it ready to use.
JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.