Skip to content

Commit

Permalink
Refactoring mesa.AgentSet (#8)
Browse files Browse the repository at this point in the history 
* updated documentation

* Update README.md

* API implementation of AgentSet and Agents

1) Agents of the same type are stored in a pd.DataFrame of the AgentSetDF class. This avoids having many missing values (which would occupy memory) if Agents or different type were stored in the same Dataframe. All agents of the model are stored in the AgentsDF class as a list of the AgentSetDF, which virtually supports the same operations as AgentSet.
2) Went with encapsulation, avoiding extensions and subclassing because they didn't always work well with storing additional attributes and it wasn't easy to extend the created classes by subclassing (as it's often done with base mesa.Agent).
3) All operations are inplace (no immutability): if we wanted to keep immutability, there couldn't be method chaining with encapsulation (since methods would have to return a pd.Dataframe) If we were to return an AgentSet. I think it also aligns well with base mesa API.
4) The "select" operations was changed to selecting "active_agents" (by a latent mask) which are the ones effectively used for the "do" operations (since usually you don't want to remove the other agents from the df but simply use a method on a part of agents of the DF). If you wanted to remove the agents you could simply use the "discard" or "remove" methods.

* fixed type annotation

* added polars backend

* dunder methods for agentsetpandas, new API, construct an abstract class

* implementation of test for methods

* distinction between abstract and concrete implementation, test implementations, update script

* Improve documentation
  • Loading branch information
@adamamer20
adamamer20 authored Jun 25, 2024
1 parent b8b7af5 commit f317ce1
Show file tree
Hide file tree
Showing 20 changed files with 3,828 additions and 721 deletions.
136 changes: 85 additions & 51 deletions README.md
View file
Original file line number Diff line number Diff line change
@@ -1,112 +1,146 @@
# mesa-frames
`mesa-frames` is an extension of the [`mesa`](https://github.com/projectmesa/mesa) designed for complex simulations with thousands of agents. By storing agents in a Pandas dataframe, `mesa-frames` significantly enhances the performance and scalability of `mesa`, while mantaining a similar syntax. `mesa-frames` allows for the use of [vectorized functions](https://vegibit.com/what-is-a-vectorized-operation-in-pandas/) whenever simultaneous activation of agents is possible.

## Why Pandas?
[Pandas](https://pandas.pydata.org/) is a popular data-manipulation Python library, developed using C and Cython. Pandas it's known both for its ease of use, allowing for declarative programming and performance. The following is a performance graph showing execution time using `mesa` and `mesa-frames` for the [Boltzmann Wealth model](https://mesa.readthedocs.io/en/stable/tutorials/intro_tutorial.html).
mesa-frames is an extension of the [mesa](https://github.com/projectmesa/mesa) framework, designed for complex simulations with thousands of agents. By storing agents in a DataFrame, mesa-frames significantly enhances the performance and scalability of mesa, while maintaining a similar syntax. mesa-frames allows for the use of [vectorized functions](https://vegibit.com/what-is-a-vectorized-operation-in-pandas/) whenever simultaneous activation of agents is possible.

![](https://github.com/adamamer20/mesa_frames/blob/main/docs/images/readme_plot.png)
## Why DataFrames?

DataFrames are optimized for simultaneous operations through [SIMD processing](https://en.wikipedia.org/wiki/Single_instruction,_multiple_data). At the moment, mesa-frames supports the use of two main libraries: Pandas and Polars.

- [Pandas](https://pandas.pydata.org/) is a popular data-manipulation Python library, developed using C and Cython. Pandas is known for its ease of use, allowing for declarative programming and high performance.
- [Polars](https://pola.rs/) is a new DataFrame library with a syntax similar to Pandas but with several innovations, including a backend implemented in Rust, the Apache Arrow memory format, query optimization, and support for larger-than-memory DataFrames.

The following is a performance graph showing execution time using mesa and mesa-frames for the [Boltzmann Wealth model](https://mesa.readthedocs.io/en/stable/tutorials/intro_tutorial.html).

![Performance Graph](https://github.com/adamamer20/mesa_frames/blob/main/docs/images/readme_plot.png)

(The script used to generate the graph can be found [here](https://github.com/adamamer20/mesa_frames/blob/main/docs/scripts/readme_plot.py))

## Installation

### Cloning the Repository
To get started with `mesa-frames`, first clone the repository from GitHub:

To get started with mesa-frames, first clone the repository from GitHub:

```bash
git clone https://github.com/adamamer20/mesa_frames.git
cd mesa_frame
cd mesa_frames
```
- ### Installing in a conda environment

### Installing in a Conda Environment

If you want to install it into a new environment:

```bash
conda create -n myenv -f requirements.txt
pip install -e .
```

If you want to install it into an existing environment:

```bash
conda activate myenv
conda install -f requirements.txt
pip install -e .
```

- ### Installing in a Python virtual environment
If you want to install into a new environment:
### Installing in a Python Virtual Environment

If you want to install it into a new environment:

```bash
python3 -m venv myenv
source myenv/bin/activate # On Windows, use `myenv\Scripts\activate`
pip install -r requirements.txt
pip install -e .
```

If you want to install into an existing environment:
If you want to install it into an existing environment:

```bash
source myenv/bin/activate # On Windows, use `myenv\Scripts\activate`
pip install -r requirements.txt
pip install -e .
```

## Usage
NOTE: `mesa-frames` is currently in its early stages of development. As such, the usage patterns and API are subject to change. Breaking changes may be introduced. Reports of feedback and issues are encouraged.

You can find the API documentation [here](https://adamamer20.github.io/mesa-frames/api)
**Note:** mesa-frames is currently in its early stages of development. As such, the usage patterns and API are subject to change. Breaking changes may be introduced. Reports of feedback and issues are encouraged.

You can find the API documentation [here](https://adamamer20.github.io/mesa-frames/api).

### Creation of the model
### Creation of an Agent

Creation of the model is fairly similar to the process in `mesa`.
A new method [`create_agents`](https://github.com/adamamer20/mesa_frames/blob/main/mesa_frames/model.py#L131) is introduced which substitutes the addition of agents to the scheduler.
The [`run_model`](https://github.com/adamamer20/mesa_frames/blob/main/mesa_frames/model.py#L71) method also accepts the number of steps which the model has run.
The agent implementation differs from base mesa. Agents are only defined at the AgentSet level. You can import either `AgentSetPandas` or `AgentSetPolars`. As in mesa, you subclass and make sure to call `super().__init__(model)`. You can use the `add` method or the `+=` operator to add agents to the AgentSet. Most methods mirror the functionality of `mesa.AgentSet`. Additionally, `mesa-frames.AgentSet` implements many dunder methods such as `AgentSet[mask, attr]` to get and set items intuitively. All operations are by default inplace, but if you'd like to use functional programming, mesa-frames implements a fast copy method which aims to reduce memory usage, relying on reference-only and native copy methods.

```python
from mesa_frames.model import ModelDF
from mesa_frames.agent import AgentDF
from mesa-frames import AgentSetPolars

class MyModel(ModelDF):
def __init__(self, N):
super().__init__()
self.num_agents = N
self.create_agents(N, {MyAgent: 1})
class MoneyAgentPolars(AgentSetPolars):
def __init__(self, n: int, model: ModelDF):
super().__init__(model)
# Adding the agents to the agent set
self += pl.DataFrame(
{"unique_id": pl.arange(n, eager=True), "wealth": pl.ones(n, eager=True)}
)

def step(self) -> None:
# The give_money method is called
self.do("give_money")

def give_money(self):
# Active agents are changed to wealthy agents
self.select(self.wealth > 0)

# Receiving agents are sampled (only native expressions currently supported)
other_agents = self.agents.sample(
n=len(self.active_agents), with_replacement=True
)

# Wealth of wealthy is decreased by 1
self["active", "wealth"] -= 1

# Compute the income of the other agents (only native expressions currently supported)
new_wealth = other_agents.group_by("unique_id").len()

# Add the income to the other agents
self[new_wealth, "wealth"] += new_wealth["len"]
```

### Creation of an agent
The agent implementation differs from base `mesa`. Agents are only defined at the class level.
The `dtypes` dictionary attribute defines the name and data types of the new columns that will be added to the MyModel.agents dataframe during the execution `create_agents`.
All methods of an `AgentDF` should be class methods and every operation within class methods should operate on the `cls.model.agents` DataFrame, filtering by the class's mask initialized during the execution of `create_agents`. The step method is defined for the entire class.
### Creation of the Model

```python
from mesa_frames.agent import AgentDF
Creation of the model is fairly similar to the process in mesa. You subclass `ModelDF` and call `super().__init__()`. The `model.agents` attribute has the same interface as `mesa-frames.AgentSet`. You can use `+=` or `self.agents.add` with a `mesa-frames.AgentSet` (or a list of `AgentSet`) to add agents to the model.

class MyAgent(AgentDF):
dtypes: dict[str, str] = {"wealth": "int64"}
```python
from mesa-frames import ModelDF

@classmethod
def __init__(cls):
class MoneyModelDF(ModelDF):
def __init__(self, N: int, agents_cls):
super().__init__()
#The next line sets the wealth attribute of all agents that are type MyAgent to 1
cls.model.agents.loc[cls.mask, "wealth"] = 1

@classmethod
def step(cls):
wealthy_agents = cls.model.agents.loc[cls.mask, "wealth"] > 0
if wealthy_agents.any():
#The next line finds the mask of a random sample of agents which is of the same length as wealthy agents
other_agents = cls.model.agents.index.isin(
cls.model.agents.sample(n=wealthy_agents.sum()).index
)
cls.model.agents.loc[wealthy_agents, "wealth"] -= 1
cls.model.agents.loc[other_agents, "wealth"] += 1
self.n_agents = N
self.agents += MoneyAgentPolars(N, self)

def step(self):
# Executes the step method for every agentset in self.agents
self.agents.do("step")

def run_model(self, n):
for _ in range(n):
self.step()
```

## What's Next?
- Refine the API to make it more understandable for someone who is already familiar with the `mesa` package. The goal is to provide a seamless experience for users transitioning to or incorporating `mesa-frames`.
- Adding support for default `mesa` functions, ensure that the standard mesa functionality is preserved.
- Adding support for or shifting to `polars` instead of pandas. This change aims to boost performance and decrease memory consumption, which is crucial for models with milions of agents.

- Refine the API to make it more understandable for someone who is already familiar with the mesa package. The goal is to provide a seamless experience for users transitioning to or incorporating mesa-frames.
- Adding support for default mesa functions to ensure that the standard mesa functionality is preserved.
- Adding GPU functionality (cuDF and Rapids).
- Creating a decorator that will automatically vectorize an existing mesa model. This feature will allow users to easily tap into the performance enhancements that mesa-frames offers without significant code alterations.
- Creating a unique class for AgentSet, independent of the backend implementation.

## License

`mesa-frames` is made available under the MIT License. This license allows you to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software, subject to the following conditions:
mesa-frames is made available under the MIT License. This license allows you to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software, subject to the following conditions:

- The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
- The software is provided "as is", without warranty of any kind.

For the full license text, see the [LICENSE](https://github.com/adamamer20/mesa_frames/blob/main/LICENSE) file in the GitHub repository.
For the full license text, see the [LICENSE](https://github.com/adamamer20/mesa_frames/blob/main/LICENSE) file in the GitHub repository.
Loading

0 comments on commit f317ce1

Please sign in to comment.