agentMET4FOF agents

class agentMET4FOF.agents.AgentBuffer(buffer_size=1000)

Buffer class which is instantiated in every agent to store data incrementally. This buffer is necessary to handle multiple inputs coming from agents. The buffer can be a dict of iterables, or a dict of dict of iterables for nested named data. The keys are the names of agents.

We can access the buffer like a dict with exposed functions such as .values(), .keys() and .items(), The actual dict object is stored in the variable self.buffer

buffer_filled(agent_from=None)

Checks whether buffer is filled, by comparing against the buffer_size.

Parameters:agent_from (str) – Name of input agent in the buffer dict to be looked up for. If agent_from is not provided, we check for all iterables in the buffer. For nested dict, this returns true for any iterable which is beyond the buffer_size.

check_supported_datatype(value)

Checks whether value is one of the supported data types.

Parameters:value (iterable) – Value to be checked. Returns:result Return type:boolean

clear(agent_from=None)

Clears the data in the buffer. if agent_from is not given, the entire buffer is removed.

agent_from : str

Name of agent

items()

Interface to access the internal dict’s items()

keys()

Interface to access the internal dict’s keys()

popleft(n=1)

Pops the first n entries in the buffer.

store(agent_from, data=None, concat_axis=0)

Stores data into self.buffer with the received message

Checks if sender agent has sent any message before If it did, then append, otherwise create new entry for it

Parameters:

• agent_from (dict | str) – if type is dict, we expect it to be the agentMET4FOF dict message to be compliant with older code otherwise, we expect it to be name of agent sender and data will need to be passed as parameter
• data – optional if agent_from is a dict. Otherwise this parameter is compulsory. Any supported data which can be stored in dict as buffering.
• concat_axis (int) – optional axis to concatenate on with the buffering for numpy arrays. Default is 0.

update(agent_from: str, data)

Overrides data in the buffer dict keyed by agent_from with value data

If data is a single value, this converts it into a list first before storing in the buffer dict.

values()

Interface to access the internal dict’s values()

class agentMET4FOF.agents.AgentMET4FOF(name='', host=None, serializer=None, transport=None, attributes=None, backend='osbrain', mesa_model=None)

Base class for all agents with specific functions to be overridden/supplied by user.

Behavioral functions for users to provide are init_parameters, agent_loop and on_received_message. Communicative functions are bind_output, unbind_output and send_output.

agent_loop()

User defined method for the agent to execute for loop_wait seconds specified either in self.loop_wait or explicitly via`init_agent_loop(loop_wait)`

To start a new loop, call init_agent_loop(loop_wait) on the agent Example of usage is to check the current_state of the agent and send data periodically

bind_output(output_agent)

Forms Output connection with another agent. Any call on send_output will reach this newly binded agent

Adds the agent to its list of Outputs.

Parameters:output_agent (AgentMET4FOF or list) – Agent(s) to be binded to this agent’s output channel

buffer_clear(agent_name=None)

Empties buffer which is a dict indexed by the agent_name.

Parameters:agent_name (str) – Key of the memory dict, which can be the name of input agent, or self.name. If one is not supplied, we assume to clear the entire memory.

buffer_filled(agent_name=None)

Checks whether the internal buffer has been filled to the maximum allowed specified by self.buffer_size

Parameters:agent_name (str) – Index of the buffer which is the name of input agent. Returns:status of buffer filled Return type:boolean

buffer_store(agent_from: str, data=None, concat_axis=0)

Updates data stored in self.buffer with the received message

Checks if sender agent has sent any message before If it did,then append, otherwise create new entry for it

Parameters:

• agent_from (str) – Name of agent sender
• data – Any supported data which can be stored in dict as buffer. See AgentBuffer for more information.

get_attr(attr)

Return the specified attribute of the agent.

Parameters:name – Name of the attribute to be retrieved.

handle_process_data(message)

Internal method to handle incoming message before calling user-defined on_received_message method.

If current_state is either Stop or Reset, it will terminate early before entering on_received_message

init_agent(buffer_size=1000, log_mode=True)

Internal initialization to setup the agent: mainly on setting the dictionary of Inputs, Outputs, PubAddr.

Calls user-defined init_parameters() upon finishing.

Inputs

Dictionary of Agents connected to its input channels. Messages will arrive from agents in this dictionary. Automatically updated when bind_output() function is called

Type:dict

Outputs

Dictionary of Agents connected to its output channels. Messages will be sent to agents in this dictionary. Automatically updated when bind_output() function is called

Type:dict

PubAddr_alias

Name of Publish address socket

Type:str

PubAddr

Publish address socket handle

Type:str

AgentType

Name of class

Type:str

current_state

Current state of agent. Can be used to define different states of operation such as “Running”, “Idle, “Stop”, etc.. Users will need to define their own flow of handling each type of self.current_state in the agent_loop

Type:str

loop_wait

The interval to wait between loop. Call init_agent_loop to restart the timer or set the value of loop_wait in init_parameters when necessary.

Type:int

buffer_size

The total number of elements to be stored in the agent buffer When total elements exceeds this number, the latest elements will be replaced with the incoming data elements

Type:int

init_agent_loop(loop_wait: Optional[int] = None)

Initiates the agent loop, which iterates every loop_wait seconds

Stops every timers and initiate a new loop.

Parameters:loop_wait (int, optional) – The wait between each iteration of the loop

init_parameters()

User provided function to initialize parameters of choice.

log_info(message)

Prints logs to be saved into logfile with Logger Agent

Parameters:message (str) – Message to be logged to the internal Logger Agent

on_received_message(message)

User-defined method and is triggered to handle the message passed by Input.

Parameters:message (Dictionary) – The message received is in form {‘from’:agent_name, ‘data’: data, ‘senderType’: agent_class, ‘channel’:channel_name} agent_name is the name of the Input agent which sent the message data is the actual content of the message

pack_data(data, channel='default')

Internal method to pack the data content into a dictionary before sending out.

Special case : if the data is already a message, then the from and senderType will be altered to this agent, without altering the data and channel within the message this is used for more succinct data processing and passing.

Parameters:

• data (argument) – Data content to be packed before sending out to agents.
• channel (str) – Key of dictionary which stores data

Returns:

Packed message data

Return type:

dict of the form {‘from’:agent_name, ‘data’: data, ‘senderType’: agent_class, ‘channel’:channel_name}.

reset()

This method will be called on all agents when the global reset_agents is called by the AgentNetwork and when the Reset button is clicked on the dashboard.

Method to reset the agent’s states and parameters. User can override this method to reset the specific parameters.

send_output(data, channel='default')

Sends message data to all connected agents in self.Outputs.

Output connection can first be formed by calling bind_output. By default calls pack_data(data) before sending out. Can specify specific channel as opposed to ‘default’ channel.

Parameters:

• data (argument) – Data content to be sent out
• channel (str) – Key of message dictionary which stores data

Returns:

message

Return type:

dict of the form {‘from’:agent_name, ‘data’: data, ‘senderType’: agent_class, ‘channel’:channel_name}.

send_plot(fig: Union[matplotlib.figure.Figure, Dict[str, matplotlib.figure.Figure]], mode: str = 'image')

Sends plot to agents connected to this agent’s Output channel.

This method is different from send_output which will be sent to through the ‘plot’ channel to be handled.

Tradeoffs between “image” and “plotly” modes are that “image” are more stable and “plotly” are interactive. Note not all (complicated) matplotlib figures can be converted into a plotly figure.

Parameters:

• fig (matplotlib.figure.Figure or dict of matplotlib.figure.Figure) – Alternatively, multiple figures can be nested in a dict (with any preferred keys) e.g {“Temperature”:matplotlib.Figure, “Acceleration”:matplotlib.Figure}
• mode (str) – “image” - converts into image via encoding at base64 string. “plotly” - converts into plotly figure using mpl_to_plotly Default: “image”

Returns:

graph

Return type:

str or plotly figure or dict of one of those converted figure(s)

set_attr(**kwargs)

Set object attributes.

Parameters:kwargs ([name, value]) – Keyword arguments will be used to set the object attributes.

shutdown()

Cleanly stop and shut down the agent assuming the agent is running.

Will let the main thread do the tear down.

step()

Used for MESA backend only. Behaviour on every update step.

stop_agent_loop()

Stops agent_loop from running. Note that the agent will still be responding to messages

unbind_output(output_agent)

Remove existing output connection with another agent. This reverses the bind_output method

Parameters:output_agent (AgentMET4FOF) – Agent binded to this agent’s output channel

class agentMET4FOF.agents.AgentNetwork(ip_addr='127.0.0.1', port=3333, connect=False, log_filename='log_file.csv', dashboard_modules=True, dashboard_extensions=[], dashboard_update_interval=3, dashboard_max_monitors=10, dashboard_port=8050, backend='osbrain', mesa_update_interval=0.1)

Object for starting a new Agent Network or connect to an existing Agent Network specified by ip & port

Provides function to add agents, (un)bind agents, query agent network state, set global agent states Interfaces with an internal _AgentController which is hidden from user

add_agent(name=' ', agentType=<class 'agentMET4FOF.agents.AgentMET4FOF'>, log_mode=True, buffer_size=1000, ip_addr=None, loop_wait=None, **kwargs)

Instantiates a new agent in the network.

Parameters:

• str (name) – with the same name. Defaults to the agent’s class name.
• AgentMET4FOF (agentType) – network. Defaults to AgentMET4FOF
• bool (log_mode) – Logger Agent. Defaults to True.

Returns:

AgentMET4FOF

Return type:

Newly instantiated agent

add_coalition(name='Coalition_1', agents=[])

Instantiates a coalition of agents.

agents(filter_agent=None)

Returns all agent names connected to Agent Network.

Returns:list Return type:names of all agents

bind_agents(source, target)

Binds two agents communication channel in a unidirectional manner from source Agent to target Agent

Any subsequent calls of source.send_output() will reach target Agent’s message queue.

Parameters:

• source (AgentMET4FOF) – Source agent whose Output channel will be binded to target
• target (AgentMET4FOF) – Target agent whose Input channel will be binded to source

connect(ip_addr='127.0.0.1', port=3333, verbose=True)

Only for osbrain backend. Connects to an existing AgentNetwork.

Parameters:

• ip_addr (str) – IP Address of server to connect to
• port (int) – Port of server to connect to

get_agent(agent_name)

Returns a particular agent connected to Agent Network.

Parameters:agent_name (str) – Name of agent to search for in the network

get_mode()

Returns:state – State of Agent Network Return type:str

set_agents_state(filter_agent=None, state='Idle')

Blanket operation on all agents to set their current_state attribute to given state

Can be used to define different states of operation such as “Running”, “Idle, “Stop”, etc.. Users will need to define their own flow of handling each type of self.current_state in the agent_loop

Parameters:

• filter_agent (str) – (Optional) Filter name of agents to set the states
• state (str) – State of agents to set

set_running_state(filter_agent=None)

Blanket operation on all agents to set their current_state attribute to “Running”

Users will need to define their own flow of handling each type of self.current_state in the agent_loop

Parameters:filter_agent (str) – (Optional) Filter name of agents to set the states

set_stop_state(filter_agent=None)

Blanket operation on all agents to set their current_state attribute to “Stop”

Users will need to define their own flow of handling each type of self.current_state in the agent_loop

Parameters:filter_agent (str) – (Optional) Filter name of agents to set the states

shutdown()

Shuts down the entire agent network and all agents

start_server_mesa()

Handles the initialisation for backend == “mesa”. Involves spawning two nested objects : MesaModel and AgentController

start_server_osbrain(ip_addr='127.0.0.1', port=3333)

Only for osbrain backend. Starts a new AgentNetwork.

Parameters:

• ip_addr (str) – IP Address of server to start
• port (int) – Port of server to start

unbind_agents(source, target)

Unbinds two agents communication channel in a unidirectional manner from source Agent to target Agent

This is the reverse of bind_agents()

Parameters:

• source (AgentMET4FOF) – Source agent whose Output channel will be unbinded from target
• target (AgentMET4FOF) – Target agent whose Input channel will be unbinded from source

class agentMET4FOF.agents.DataStreamAgent(name='', host=None, serializer=None, transport=None, attributes=None, backend='osbrain', mesa_model=None)

Able to simulate generation of datastream by loading a given DataStreamMET4FOF object.

Can be used in incremental training or batch training mode. To simulate batch training mode, set pretrain_size=-1 , otherwise, set pretrain_size and batch_size for the respective See DataStreamMET4FOF on loading your own data set as a data stream.

agent_loop()

User defined method for the agent to execute for loop_wait seconds specified either in self.loop_wait or explicitly via`init_agent_loop(loop_wait)`

To start a new loop, call init_agent_loop(loop_wait) on the agent Example of usage is to check the current_state of the agent and send data periodically

init_parameters(stream=<agentMET4FOF.streams.DataStreamMET4FOF object>, pretrain_size=None, batch_size=1, loop_wait=1, randomize=False)

Parameters:

• stream (DataStreamMET4FOF) – A DataStreamMET4FOF object which provides the sample data
• pretrain_size (int) – The number of sample data to send through in the first loop cycle, and subsequently, the batch_size will be used
• batch_size (int) – The number of sample data to send in every loop cycle
• loop_wait (int) – The duration to wait (seconds) at the end of each loop cycle before going into the next cycle
• randomize (bool) – Determines if the dataset should be shuffled before streaming

reset()

This method will be called on all agents when the global reset_agents is called by the AgentNetwork and when the Reset button is clicked on the dashboard.

Method to reset the agent’s states and parameters. User can override this method to reset the specific parameters.

class agentMET4FOF.agents.MesaModel

A MESA Model

step()

Advance the model by one step.

class agentMET4FOF.agents.MonitorAgent(name='', host=None, serializer=None, transport=None, attributes=None, backend='osbrain', mesa_model=None)

Unique Agent for storing plots and data from messages received from input agents.

The dashboard searches for Monitor Agents’ memory and plots to draw the graphs “plot” channel is used to receive base64 images from agents to plot on dashboard

memory

Dictionary of format {agent1_name : agent1_data, agent2_name : agent2_data}

Type:dict

plots

Dictionary of format {agent1_name : agent1_plot, agent2_name : agent2_plot}

Type:dict

plot_filter

List of keys to filter the ‘data’ upon receiving message to be saved into memory Used to specifically select only a few keys to be plotted

Type:list of str

init_parameters(plot_filter=[], custom_plot_function=-1, *args, **kwargs)

User provided function to initialize parameters of choice.

on_received_message(message)

Handles incoming data from ‘default’ and ‘plot’ channels.

Stores ‘default’ data into self.memory and ‘plot’ data into self.plots

Parameters:message (dict) – Acceptable channel values are ‘default’ or ‘plot’

reset()

This method will be called on all agents when the global reset_agents is called by the AgentNetwork and when the Reset button is clicked on the dashboard.

Method to reset the agent’s states and parameters. User can override this method to reset the specific parameters.

update_plot_memory(message)

Updates plot figures stored in self.plots with the received message

Parameters:message (dict) – Standard message format specified by AgentMET4FOF class Message[‘data’] needs to be base64 image string and can be nested in dictionary for multiple plots Only the latest plot will be shown kept and does not keep a history of the plots.