Machine-Learning-Pipeline

This repository was made for providing basic machine learning process and pipeline system. I made machine learning process by using scikit-learn. I hope you can get help from this repo.

Environments

Requirment

scikit-learn == 0.22.1

ML task

• Regression
• Classification

ML in scikit-learn

Models

Linear Models

• Linear Regression
• Logistic Regression
• Ridge
• Lasso
• ElasticNet

Tree-based Models

• Decision Tree
• Random Forest
• Adaboost
• Gradient Tree Boosting

Support Vector Machine

• SVC (classification)
• SVR (regression)

Naighbors

• K-Nearest Neighbors

Evaluation

Regression

• Mean Squared Error, MSE
• Mean Absolued Error, MAE

Classification

• Accuracy
• Recall
• Precision
• F1-score
• AUC

Save & Load

• pickle

Process

Basic process for machine learning (ML)

1. Load Data
2. Preprocessing Data
3. Model Building
4. Model Training
5. Evaluation
6. Save

Let's try coding with examples

• Regression Notbook
• Classification Notbook

Pipeline

Scripts

.
├── dataload.py
├── evaluate.py
├── main.py
├── model.py
├── train.py
└── utils.py

main.py

Argument

Machine Learning Pipeline

optional arguments:
  -h, --help            show this help message and exit
  --seed SEED           Set seed
  --task {regression,classification}
                        Choice machine learning task
  --datagen DATAGEN     Generate train and test set
  --datadir DATADIR     Set data directory
  --logdir LOGDIR       Set log directory
  --val_size VAL_SIZE   Set validation size
  --kfold KFOLD         Number of cross validation
  --modelname {OLS,Logistic,Ridge,Lasso,ElasticNet,DT,RF,ADA,GT,SVM,KNN}
                        Choice machine learning model

Code

# 1. load data
train, test = dataloader(task=args.task, datadir=args.datadir)

# 2. preprocessing
x_train, y_train = preprocessing(data=train, data_type='train')
x_test = preprocessing(data=test, data_type='test')

# 3. model setting
model = SklearnModels(task=args.task, modelname=args.modelname, random_state=args.seed)

# 4. training
if args.kfold:
    cross_validation(K=args.kfold, model=model, train=[x_train, y_train], test=x_test, args=args)
else:
    training(model=model, train=[x_train, y_train], test=x_test, args=args)

model.py

class SklearnModels:
    def __init__(self, **kwargs):
        pass

    def build(self, **kwargs):
        pass

    def fit(self, **kwargs):
        pass

    def eval(self, **kwargs):
        pass
        
    def predict(self, **kwargs):
        pass

    def load(self,, **kwargs):
        pass

dataload.py

def data_generator(**kwargs):
    pass

def dataloader(**kwargs):
    pass

def preprocessing(**kwargs):
    pass

train.py

def training(model, train: list, test, args):
    pass

def cross_validation(K, model, train: list, test, args):
    pass

evaluate.py

def evaluation_all(task, y_true, y_pred, threshold=0.5):
    pass

utils.py

def results_save(**kwargs):
    pass

def results_comparison(**kwargs):
    pass

def results_bar(**kwargs):
    pass