Bias variance tradeoff Flashcards
Error
Error = y-ŷ
Difference between the actual output - predicted output
Actual o/p (how to denote it)
In supervised machine learning, we train the model with a labeled data means i/p data with a respective o/p value (actual o/p). Denoted as ‘y’
Predicted output (how to denote it)
Once model is ready, we test the model with a testing data set. So model gives the predicted o/p denoted as ‘ŷ’
Bias
Training error
Difference between the actual output and the predicted output of a model when it is evaluated with a training data set is called as bias error
Bias error = Actual o/p - Predicted o/p w.r.t. training dataset
Variance
Testing error
Difference between the actual o/p and the predicted o/p of a model w.r.t the testing data (unseen data) is called as variance error.
Variance error = actual o/p - predicted o/p w.r.t. training dataset
Overfitting
Model type also
Low bias, High Variance
Too much training results in a model that performs well on the training data but poorly on new unseen data (tested data).
Complex model: In overfitting, the model is often too complex, with too many parameters relative to the number of observations.
Underfitting
Model type also
High bias, high variance
Model fails to learn the training data and therefore performs poorly on both the training and test datasets.
Simple model: The model is often too simple, with few parameters relative to the complexity of the data.
Graph of bias variance trade off
Explanation of (Graph of bias variance trade off)
Summary of bias variance tradeoff
Graphical representation according to statistical analysis of real life projects with complexity of models (graph 1st)
Graphical representation according to statistical analysis of real life projects with complexity of models (graph 2nd)
Graphical representation according to statistical analysis of real life projects with complexity of models (graph 3rd)
