MACHINE LEARNING
MANIKANTA
Machine Learning job is done by half machine and half human.
Why Machine Learning?
TP VS ML
STEP 1:
Training Data
Training Testing Validation
Input features Only Input features Only Input features
Labels No Labels No Labels
Model Learns on this Dataset Tests the model performance Tests the model peroformance
50% of the Data 20% of the Data 30% of the data
Model is predicted based on Model is reconstructed based Determines the accuracy of
the data. on the outputs of testing data the model
STEP 2:
Model
Model Provides the mathematical mapping between input features and output labels.
Output = weight0 + weight1 * feature1 + weight2 * feature2 +· · · + weightm * featurem
The Input is a bunch of features and output can be a real number or Discrete value from a finite
set.
If output is real number: Regression
If output is discrete value: Classification
STEP 3:
Loss Function
We have to predict a nice model for the use.
Then we need a suitable method for measuring the difference between actual and predicted
output.
Loss function provides that measure.
Loss function is a function of weight vector, as we change the weight vector we will get different
loss.
J(w)=Σ(predicted – actual)2
STEP 4:
Optimization
Obtain the weight vector that minimizes the loss function.
d J(w)=0
dw
Solving the above equation we get the weight vector.
Some methods are:
Normal Equation
Gradient Descent
Mini Batch Gradient Descent
Stochastic Gradient Descent
STEP 5:
Evaluation
Evaluation validates the function.
The evaluation metric changes as per the problem – different metric for regression and
classification problems.
Regression: Classification:
Mean Squared Error (MSE) Accuracy, Precision, and Recall
Root Mean Squared Error (RMSE) F1 Score
Mean Absolute Error (MAE) Log Loss/Binary Crossentropy
Categorical Crossentropy
AUC
THANK YOU
Source: IIT MADRAS
Credit by : @manikanta_puppala, 2022.
