Skip to main content

Semantic Learning algorithm based on Inflate and deflate Mutation (SLIM GSGP)

Project description

SLIM (Semantic Learning algorithm based on Inflate and deflate Mutation)

gsgp_slim is a Python library that implements the SLIM algorithm, which is a variant of the Geometric Semantic Genetic Programming (GSGP). This library includes functions for running standard Genetic Programming (GP), GSGP, and all developed versions of the SLIM algorithm. Users can specify the version of SLIM they wish to use and obtain results accordingly. Slim's documentation can be accessed in Slim Documentation.

Installation

To install the library, use the following command:

pip install slim_gsgp

Usage

Running GP

To use the GP algorithm, you can use the following example:

from slim_gsgp.main_gp import gp  # import the slim_gsgp library
from slim_gsgp.datasets.data_loader import load_ppb  # import the loader for the dataset PPB
from slim_gsgp.evaluators.fitness_functions import rmse  # import the rmse fitness metric
from slim_gsgp.utils.utils import train_test_split  # import the train-test split function

# Load the PPB dataset
X, y = load_ppb(X_y=True)

# Split into train and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, p_test=0.4)

# Split the test set into validation and test sets
X_val, X_test, y_val, y_test = train_test_split(X_test, y_test, p_test=0.5)

# Apply the GP algorithm
final_tree = gp(X_train=X_train, y_train=y_train,
                X_test=X_val, y_test=y_val,
                dataset_name='ppb', pop_size=100, n_iter=100)

# Show the best individual structure at the last generation
final_tree.print_tree_representation()

# Get the prediction of the best individual on the test set
predictions = final_tree.predict(X_test)

# Compute and print the RMSE on the test set
print(float(rmse(y_true=y_test, y_pred=predictions)))

Running standard GSGP

To use the GSGP algorithm, you can use the following example:

from slim_gsgp.main_gsgp import gsgp  # import the slim_gsgp library
from slim_gsgp.datasets.data_loader import load_ppb  # import the loader for the dataset PPB
from slim_gsgp.evaluators.fitness_functions import rmse  # import the rmse fitness metric
from slim_gsgp.utils.utils import train_test_split  # import the train-test split function


# Load the PPB dataset
X, y = load_ppb(X_y=True)

# Split into train and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, p_test=0.4)

# Split the test set into validation and test sets
X_val, X_test, y_val, y_test = train_test_split(X_test, y_test, p_test=0.5)

# Apply the Standard GSGP algorithm
final_tree = gsgp(X_train=X_train, y_train=y_train,
                  X_test=X_val, y_test=y_val,
                  dataset_name='ppb', pop_size=100, n_iter=100,
                  reconstruct=True, ms_lower=0, ms_upper=1)

# Get the prediction of the best individual on the test set
predictions = final_tree.predict(X_test)

# Compute and print the RMSE on the test set
print(float(rmse(y_true=y_test, y_pred=predictions)))

Running SLIM

To use the SLIM GSGP algorithm, you can use the following example:

from slim_gsgp.main_slim import slim  # import the slim_gsgp library
from slim_gsgp.datasets.data_loader import load_ppb  # import the loader for the dataset PPB
from slim_gsgp.evaluators.fitness_functions import rmse  # import the rmse fitness metric
from slim_gsgp.utils.utils import train_test_split  # import the train-test split function
from slim_gsgp.utils.utils import generate_random_uniform  # import the mutation step function

# Load the PPB dataset
X, y = load_ppb(X_y=True)

# Split into train and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, p_test=0.4)

# Split the test set into validation and test sets
X_val, X_test, y_val, y_test = train_test_split(X_test, y_test, p_test=0.5)

# Apply the SLIM GSGP algorithm
final_tree = slim(X_train=X_train, y_train=y_train,
                  X_test=X_val, y_test=y_val,
                  dataset_name='ppb', slim_version='SLIM+SIG2', pop_size=100, n_iter=100,
                  ms_lower=0, ms_upper=1, p_inflate=0.5)

# Show the best individual structure at the last generation
final_tree.get_tree_representation()

# Get the prediction of the best individual on the test set
predictions = final_tree.predict(X_test, slim_version='SLIM+SIG2')

# Compute and print the RMSE on the test set
print(float(rmse(y_true=y_test, y_pred=predictions)))

Arguments for the gp, gsgp and slim function

Common arguments

  • X_train : A torch tensor with the training input data (default: None).
  • y_train : A torch tensor with the training output data (default: None).
  • X_test : A torch tensor with the testing input data (default: None).
  • y_test : A torch tensor with the testing output data (default: None).
  • dataset_name : A string specifying how the results will be logged (default: None).
  • pop_size : An integer specifying the population size (default: 100).
  • n_iter : An integer specifying the number of iterations (default: 1000).
  • elitism : A boolean specifying the presence of elitism (default: True).
  • n_elites : An integer specifying the number of elites (default: 1).
  • init_depth : An integer specifying the initial depth of the GP tree
    • default: 6 for gp and slim
    • default: 8 for gsgp
  • log_path : A string specifying where the results are going to be saved
    • default: os.path.join(os.getcwd(), "log", "gp.csv") for slim
    • default: os.path.join(os.getcwd(), "log", "gsgp.csv") for slim
    • default: os.path.join(os.getcwd(), "log", "slim.csv") for slim
  • seed: An integer specifying the seed for randomness (default: 1).

Specific for gp

  • p_xo : A float specifying the crossover probability (default: 0.8).
  • max_depth : An integer specifying the maximum depth of the GP tree (default: 17).

Specific for gsgp

  • p_xo : A float specifying the crossover probability (default: 0.0).
    • ms: A callable function to generate the mutation step (default: generate_random_uniform(0, 1)).

Specific for slim

  • slim_version: A string specifying the version of SLIM-GSGP to run (default: "SLIM+SIG2").
  • ms: A callable function to generate the mutation step (default: generate_random_uniform(0, 1)).
  • p_inflate: A float specifying the probability to apply the inflate mutation (default: 0.5).

Additional Notes

If a user wishes to use their own dataset rather than one of the sixteen benchmarking datasets included with the slim library, they can load their data into a Pandas DataFrame, ensuring that the target variable is the last column. They can then call the load_pandas_df function from datasets.data_loader as follows:

from slim_gsgp.datasets.data_loader import load_pandas_df  # import the loader for the dataset PPB
import pandas as pd

# Reading the desired dataset
df = pd.read_csv("path/your_data.csv")

# Turning df into X and y torch.Tensors
X, y = load_pandas_df(df, X_y=True)

# Split into train and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, p_test=0.4)

# Split the test set into validation and test sets
X_val, X_test, y_val, y_test = t

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

slim_gsgp-0.1.3.tar.gz (56.7 kB view details)

Uploaded Source

Built Distribution

slim_gsgp-0.1.3-py3-none-any.whl (101.4 kB view details)

Uploaded Python 3

File details

Details for the file slim_gsgp-0.1.3.tar.gz.

File metadata

  • Download URL: slim_gsgp-0.1.3.tar.gz
  • Upload date:
  • Size: 56.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.1 CPython/3.10.15

File hashes

Hashes for slim_gsgp-0.1.3.tar.gz
Algorithm Hash digest
SHA256 0e0c7a5bbc106ce97e77bae1f11a2ff167fd70405841e7e7a1b7acf2282bcb14
MD5 6432bdb8b753d4c8dd87812b4a4f2580
BLAKE2b-256 3d7d49baab3e555564f925be316db2dba2de7d0a9e23c671302691d24ff2a910

See more details on using hashes here.

File details

Details for the file slim_gsgp-0.1.3-py3-none-any.whl.

File metadata

  • Download URL: slim_gsgp-0.1.3-py3-none-any.whl
  • Upload date:
  • Size: 101.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.1 CPython/3.10.15

File hashes

Hashes for slim_gsgp-0.1.3-py3-none-any.whl
Algorithm Hash digest
SHA256 f132f0079e6734bb7a30b5da11f0045f59b0e1666d2e0c893e7d80736bfb7d6c
MD5 ec6bd90bd069d14a6cabd99e6877185f
BLAKE2b-256 5de493e1060955a109a050bf0dc59783d07d5d6baa4df522237c0caff03f1c90

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page