Module 2

Machine Learning Pipeline

• Framing a machine learning problem
• Get the data
• Data Pre processing
• Data Visualization and Analysis
• Feature Engineering
• Model Building and Evaluation
• Fine tune the model
• Present the solution
• Launch, monitor and maintain

Case Study

Bank is loosing too much money due to bad loans and wants to reduce losses. The following things are looked at to frame the problem

Frame the problem

• Define the objectives in business terms
• How will your solution be used
• What are the current solutions
• How should you frame this problem (Supervised or unsupervised)
• How should the performance be measured
• Is the performance measure aligned with the business objective
• What are comparable problems
• Is Human expertise available
• How would you solve the problem?
• List the assumptions you have made
• What would be the minimum performance needed to reach the business objective?

An example problem statement

• Build a model of housing prices using census data
  • Data attributes and so on for each district
  • Districts are the smallest geographical unit (population ~600-3000)
• The model to predict the median housing price in an district, given all the other metrics
• Goodness of the model is determined by how close the model output is wrt actual price for unseen data.

Framing the problem

• What is the expected usage and benefit?
  • Impacts the choice of algorithms, goodness measure, and effort in lifecycle management of the model
• What is the baseline method and its performance?

• Analyze the dataset

  • Each instance comes with the expected output
  • Hence we are going with supervised
  • Goal is to predict a real valued price based on multiple variables line population, income etc
    • Regression is chosen for this reason
  • Output is based on input data at rest, not rapidly changing data rapidly.
  • Dataset small enough to fit in memory
    • Batch
  • Since we are predicting a single value we are calling it a univariate problem

• Choice of Performance metrics

  • Root Mean Square
    
  • Mean Absolute Error (MAE)
    

• Types and properties of attributes
  
  

Data types of attributes

job, marital -> nominal
education -> Ordinal
credit default -> Binary (symmetric)
housing loan -> Binary

• Categorical
• Binary
  • Symmetric (Simple matching coefficient proximity measure)
  • Asymmetric (Jaccart coefficient proximity measure)
• Nominal
• Numerical/continuous (Minkowsky distance for proximity measure)
• to find outliers

Some pandas functions

df.types() # Get the datatypes of all columns in dataframe
df.info() # Get more info on non null values in data frame
df.describe() # Get statistical summary of each attribute and 5 point summary of numberic attributes

Data Types

• relational/Object data
• Transactional data
• Document data
• Web and social network data
• Spatial data
• time series data

Note: Topics such as data types, pre processing, visualization and analysis are explained well in Jupyter Notebooks

---------------------------------

Sampling

Binarization

One Hot Encoding

Attribute Transformation

- This is done to reduce skewness in data.
- Transformation is done to bring data to the same scale that helps in feature scaling

Normalization

• 
• 
• 
  
  Robust scaler normalizes withing the IQR (Q3 - Q1)

Curse of Dimensionality

Dimensionality Reduction

- PCA:
- Other ways to reduce dimensionality:
- Correlation Decision:
- Chi Square Stats (Categorical attributes)
- Pearson Coefficient (Numeric attributes)
- Annova Test (Mixed attributes)
- Wrapper Method is to train models with several subsets of data and choose the one that was most fruitful do then do further improvement on that.

Feature Creation

---

Tags: !AMLIndex

---

Tags: !AMLIndex