pyecoacc

A python package for supervised learning of behavioral modes from sensor data.

pyecoacc is built on top of sklearn and pytorch, and provides convenient pipelines of feature computations and other preprocessing necessary to easily run behavior classification on sensor data. This package contains easy to use out of the box optoins, together with extendability and customization.

Documentation

You can find the full documentation here. Complete workflow examples can be found in the examples folder.

📦 Installation

Install from PyPI:

pip install pyecoacc   

Or from the source:

git clone https://github.com/Hezi-Resheff/pyecoacc.git
cd pyecoacc
pip install .

🔗 Dependencies

All required dependencies are listed in requirements.txt. To install them manually:

pip install -r requirements.txt

🧑‍💻 Usage

pyecoacc allows a wide range of classifiers ranging from simple out of the box models, to custom pipelines and deep learning.

Simple out of the box classifier:

from pyecoacc.models.pipeline import get_default_random_forest_pipeline

classifier = get_default_random_forest_pipeline()
classifier.fit(ACC_train, y_train)
y_hat = classifier.predict(ACC_test)

Simple XGBoost based model:

from xgboost import XGBClassifier
from pyecoacc.models.pipeline import make_classifier_pipeline

xg_model = make_classifier_pipeline(features=ACCStatsTransformer(), 
                                    model=XGBClassifier(n_estimators=250)) 

Custom CNN classifier (don't actually use anything so small unless it's just as a quick baseline...):

from pyecoacc.models.deep.cnn import make_cnn_model


tiny_cnn_model = make_cnn_model(input_dim=INPUT_ACC_LENGTH, 
                                num_behav=NUMBER_OF_BEHAVIORS, 
                                conv_filters=[32, 64],
                                fc_layers=[20],
                                verbose=0)

classifier.fit(ACC_train, y_train)
y_hat = classifier.predict(ACC_test)