Extend logging capabilities and agent configurability

[Archento]

trafficstars

We need to improve logging throughout the whole framework and enable the user to configure verbosity as well as output of the logs.

• agent already has log level as property so we'll only need to provide the means for piping the logs to a file (or similar means)
• logging within the framework is mainly happening on the INFO level but under-the-hood actions should be logged as well to properly debug a running agent

Potential solution could be a config mapping which may be created by the user or generated and filled with default values on demand by running the agent with a CLI argument.

Agent instantiation

agent = Agent(
    name="<your_name>",
    seed="<your_seed>",
    [...],
    config=AgentConfig(
        path="./my_agent.yaml",
        **kwargs   # maybe individual config values
    )
)

my_agent.yaml

name: "bert"
port: 8000
seed: "..."
endpoint: "http://localhost:8000/submit"
enable_wallet_messaging: false
version: "v20.0.9"
log_level: "DEBUG"
log_file: "./${timestamp}_${agent_address}.log"
...

---

By using this approach it should also be easier to scale agent deployment if we can put every agent init argument inside the config map, enabling the developer to invoke:

configs= ["./config_a.yaml", "./config_b.yaml", ...]  # list of config files

agents = []
for conf in configs:
    agents.append(Agent(config=conf))

[Archento]

Add automatic logging of agent address on agent startup.

[Archento]

I think there are two main approaches when it comes to configuration files:

• simply loading the file (.yaml in this case), create a dict of values to be used as a direct way of creating an agent, or
• loading an AgentConfig object from a file which includes type validation to create agents from.

Both can coexist but we may want to force one of the usage of one over the other.

---

We can think about adding the AgentConfig object to the agent by default so that config handling will be more natively embedded in the framework.

---

Using yaml files makes handling of multiple agents very easy as the configs just need to be separated by --- for it to be loaded like this:

"""yaml contents
name: alice
seed: ...
---
name: bob
seed: ...
"""

confs = load_configs(PATH) # multiple configs in one file

bureau = Bureau()
for conf in confs:
    bureau.add(Agent(**conf))

bureau.run()

---

Requirements

• after the process, one (or multiple of) working agent(s) needs to be instantiated
• the process should feel native, i.e. nothing should fundamentally change from a dev perspective apart from slight additions