This slide is a part of the Introductory Machine Learning Lab Sessions that I'm currently conducting in this Spring 2018 semester at Daffodil International University. I hope this would be helpful for my students as well as other enthusiasts.
2. Contents
What is Machine Learning
Supervised, Unsupervised, Reinforcement
Learning
Training, Test, Validation data
Basic Workflow of any ML Project
Cognota.ai
13. Train,Validation,
Test
(Tutorial)
Cognota.ai
Data
1st Split Data Test Data
Training Data Validation Data
Training (60%) Validation (20%) Test (20%)
Train your model to
fit the parameters
Tune the hyper
parameters of your
model, avoid over
fitting, choose
model
Test your model to
determine accuracy
and performance
15. Cause
&
Prevention
AVOID OVER FITTING
Happens when number of features is much higher than
number of training examples
Increase number of training data (sometimes might not
work)
Use Regularization (will talk later about this)
AVOID UNDER FITTING
When number of features is much lower than number of
training examples
Increase number of features
Might use compound features, or can go to a higher
dimensional feature space using Kernels (will discuss later)
Cognota.ai
16. Machine
LearningProject
Framework
Cognota.ai
1. Data Import
2. Data Preprocessing
3. Feature Engineering and Extraction
4. Model Selection
5. Train Model with Training Data
6. Tune the Hyper Parameters with Validation
Data (learning rate, regularization parameter,
number of layers in Neural Network etc.)
7. Execute Model on Test Data
8. Performance Evaluation
9. Go Back to step 3 and repeat and until
satisfactory accuracy achieved
10.Conclude