keras ensure equal class representation during traingin
from sklearn.utils import class_weight
class_weights = class_weight.compute_class_weight('balanced',
np.unique(y_train),
y_train)
model.fit(X_train, y_train, class_weight=class_weights)