easytrainer 0.1.4

A package to facilitate your model training

easytrainer

Prepare your data and train your models easily.

Works with scikit-learn.

Installation

pip install easytrainer

Usage

Preparator

---

Data preparators integrate preparation methods. When instantiating the preparator, simply choose the order of execution for each method and fill in the parameters if needed.

from easytrainer.data.text import TextualPreparator

preparator = TextualPreparator(
    to_lower=1,
    drop_stopwords=2
)

easytrainer.data.text.TextualPreparator

Method Name	Description	Type
to_lower	Order in which to apply lowercase transformation	(Optional[int])
to_upper	Order in which to apply uppercase transformation	(Optional[int])
drop_stopwords	Order in which to remove French stopwords	(Optional[int])
drop_digits	Order in which to remove digits	(Optional[int])
lemmatize	Order in which to apply lemmatization using spaCy	(Optional[int])
drop_special_characters	Order to remove special characters (punctuation etc.)	(Optional[int])
drop_accents	Order in which to remove accents from characters	(Optional[int])
drop_words_less_than_N_letters	Either the order (int), or a tuple (order, n_letters, isalpha_flag) indicating the order, minimum number of letters, and whether to filter only alphabetic words	(Optional[Union[int, Tuple[int, int, bool]]])

Coming soon: easytrainer.data.numeric easytrainer.data.image easytrainer.data.video easytrainer.data.audio

Then call prepare() to preprocess your data:

data = [
    "J'adore cette musique, elle est magnifique.",
    "Je déteste ce film, il est ennuyeux."
]

results = preparator.prepare(data)
print(results["data"])
>>> ["j'adore musique, magnifique.", 'déteste film, ennuyeux.']

TextualPreparator.prepare()

Parameters	Description	Type
data	Text data to preprocess. If a numpy array is provided, it will be converted to a Series (1D) or DataFrame (2D)	(List[str] or pd.Series or pd.DataFrame or np.ndarray)
all	If True and data is a DataFrame, apply transformations to all columns (converted to str). If False, apply only to object/string columns	bool (default : False)
encoder_name_to_fit	If specified, encode the processed data using this method ('tfidf', 'word2vec')	(Optional[str])
custom_encoder_to_fit	A user-provided encoder with a fit_transform() method	(Optional[object])
custom_encoder_fit	A user-provided fitted encoder with a transform() method	(Optional[object])

Returns a dictionary:

• If no encoder: a dict containing 'data' (the preprocessed data).
• If encoder is specified: a dict containing 'data', 'encoded_data', and 'vectorizer'.

Models

---

Easily train and evaluate your models.

from sklearn.ensemble import RandomForestClassifier

from easytrainer.models.classifier import SklearnClassifierModel

SCM = SklearnClassifierModel(
    model=RandomForestClassifier(n_estimators=100, random_state=42),
)

# Training
SCM.fit(X_train, y_train, param_grid={'n_estimators': [50, 100]}, cv=3)

# Testing
report = SCM.test(X_test, y_test)
print(report)

# Validation
validation_results = SCM.validation(X_val, y_val, fallback_class=0)

# Prediction
predictions = SCM.predict(X_test, threshold=0.7, fallback_class=0)

easytrainer.models

Method Name	Description	Args	Returns
fit	Trains the model, with optional hyperparameter tuning using GridSearchCV.	(X, y, param_grid=None, **kwargs)	self.best_fit_model (fitted model)
test	Evaluates the model on a test set, returning a classification report.	(X, y)	pd.DataFrame containing precision, recall, f1-score, support per class
evaluate	Evaluates model performance across different probability thresholds, optionally using a fallback class if prediction confidence is too low.	(X, y, fallback_class, validation_thresholds=None)	pd.DataFrame with scores (accuracy, weighted/macro F1, per-threshold metrics)
predict	Makes predictions, optionally applying a threshold and fallback class.	(X, threshold=None, fallback_class=None)	np.ndarray (predicted labels)
save	Saves the fitted model to disk.	(path)	None
load	Loads a previously saved model from disk.	(path)	SklearnClassifierModel instance with loaded model