ESBM - EarlyStopping customized to your own needs = metrices (including threshold optimization and best model serailization)
This package will be assiting easily you with several training phases:
- First, you should define your required metric.
- ESBM will then evaulate your model performance on the validation set.
- It will select the best classsification threshold and inform you about the results.
- It will save the best evaualted model for future use.
- It will initate early stopping after a defined period with no metric improvment.
First you got to initialize your earlystopping object.
from earlyStopping import EarlyStoppingByMetric
ESBM = EarlyStoppingByMetric(x_val, y_val, patience = 5, batch_size = 256)Then you just use it in your .fit as a another callback
model.fit(Xtr, Ytr, validation_data = (Xv,Yv),epochs=50, batch_size=256, verbose=1,callbacks=[ESBM],shuffle=True)
There are a few more arguments you are able to pass into the ESBM object in order to recieve your ideal results.
ESBM = EarlyStoppingByMetric(x_val, y_val, patience, batch_size, threshold_searching = 50, metric = "precision", min_samples = 50)How long would you like to wait before earlystopping initiation?
ESBM = EarlyStoppingByMetric(... , patience = 10 , ...)How many iterations to perform while looking for your best classification threshold?
ESBM = EarlyStoppingByMetric(... , threshold_searching = 50 , ...)Which metric which you like to optimize?
ESBM = EarlyStoppingByMetric(... , metric = "precision" , ...)What is the minimum amount of samples would you like to take into account while optimizing metric on validation set?
ESBM = EarlyStoppingByMetric(... , min_samples = 50 , ...)