Link
Machine Learning

This site uses Just the Docs, a documentation theme for Jekyll.

Search

Fundamentals of Machine Learning

1. Define Machine Learning
Machine Learning is about building systems that can learn from data.
Learning means getting better at some task, given some performance measuermeasure.

2. List out four types of problems that can be solved using it.
Machine Learning is great for…

  • Problems for which existing solution require a lot of fine-tuning or long list of ruls.
    One Machine Learning algorithm can often simplify code and perform better than the traditional approach.
  • Complex problem for which using a traditional approach yields no good solution.
    The best Machine Learnign techniques can perhaps find a solution.
  • Fluctuating environments: A Machine Learning system can adapt to new data.
  • Getting insights about complex problems and large amounts of data.(e.g., data mining)

Types of Machine Learning Systems

There are so many different types of Machine Learning systems that it is useful to classify them in broad categories, based on the following criteria:

  • Whether or not they are trained with human supervision
    (supervised, unsupervised, semisupervised and reinforcement learning)
  • Whether or not they learn incrementally on the fly (online versus batch learning)
  • Insance based(work by simply comparing new data points to known data points) OR Model based(work by detecting patterns in the training data and building a predictive model)

3. What is a labeled training set?
Labeled training set is a training set which contains the label(desired solution) for each instance.

4. What are the two most common supervised tasks?
The most common supervised tasks are Classification and Regression.

5. What are the four most common unsupervised tasks?

  • Clustering
  • Anomaly detection and novelty detection
  • Visualization and dimensionality reduction
  • Association rule learning

6. What type of Machine Learning algorithms would you use to allow a robot to walk in various unknown terrains?
Reinforcement Learning

7. What type of algorithm would you use to segment your customers into multiple groups?
If we can defien the lables then, classification algorithms(supervised learning) But if we can not define (or identify) the labels then, clustering algorithms (unsupervised learning)

8. Spam detection is in example of Supervised learning Or Unsupervised learning?
Spam detection is a supervised learning problem.

9. What is an online learning system?
In online learning, you train the system incrementally by feeding it data instances sequentially, either individually or in small groups called mini-batches.

  • Learning is fast and cheap, so the system can learn about new data on the fly, as it arrives.
  • Great for system that receive data as continuous flow, And need to adapt to change rapidly.

10. What is out-of-core learning?
An Out-of-core learning algorithm splits the data into mini-batches and uses online learning(incremental) algorithms to learn from these mini-batches.This is useful when the entire data can not fit into computer’s memory.

11. What type of learning algorithm relies on a similarity measure to make predictions?
Instance based learning algorithm relies on a similarity measure to make predictions.

12. What is the difference between a model parameter and a learning algorithm’s hyperparameter?
A model has one or more model parameter that determines what it will predict given a new instance. A learning algorithm tries to find optimal values for these parameters such that the model generalizes well to new instances

A hyperparameter is a parameter of the learning algorithm itself, not of the model. (e.g., the amount of regularization to apply).

13. What do model-based learnign algorithms search for? What is the most commmon strategy they use to succeed? How do they make predictions?

  • Modle based learning algorithms search for an optimal value for the model parameter such that the model will generalize well to new instances.
  • The model gets trained by minimizing a cost function that measures how bad the system is at making predictions on the training data, plus a penalty for model complexity if the model is regularized.
  • The trained Modles make predictions by taking features of new instance and predicting using the learned function.

14. Can you name four main challenges in Machine Learning?

Challenges wrt Bad Data

  • Insufficient quantity of training data: It takes a lot of data for most ML algorithms to work properly.
  • Nonrepresentative Training Data: IT is crucial to have training data that is representative of the new cases. 
     If the sample is too small, you will have sampling noise, but even very large samples can be nonrepresentative if the sampling method is flawed. This is called sampling bias.
  • Poor-Quality data: If training data is full of errors, outliers, and noise, it will make harder for the system to detect the underlying patterns. 
     Most data scientists spend a significant part of their time doing data cleaning only.
  • Irrelevant Features: System will only capable of learning if the training data contains enough relevant features and not too many irrelevant ones. 
     Feature engineering: A critical part of the success of a Machine Learning project is coming up with a good set of features to train on.

Challenges wrt Bad Algorithms

  • Overfitting the Training Data: The model performs well on the training data, but it does not generalize well.
  • Underfitting the Training Data: The mode is too simple to learn the underlyting structure of the data.

15. If your model performs great on the training data but generalizes poorly to new instances, What is happenning? Can you name three possible solutions?
Here, The model is overfitt to the data. That happens when the model is too complex relative to the amount and noisiness of the training data. Followings are the possible solutions:

-Simplifying the model by selecting one with fewer parameters, which is by choosing linear model over high-degree polynomial model(reduce the number of attributes in the training data).

  • Gather more training data.
  • Reduce the noise in the training data, like by fixing data errors and remove outliers. 
     Constraining a model to make it simpler and reduce the risk of overfittign is called regularization.

16. What is Underfitting? And How to fix it?
Underfitting occurs when the model is too simple to learn the underlying structure of the data. Followings are the possible fix:

  • Select a more powerful model with more parameters.
  • Feed better features to the learning algorithm (feature engineering).
  • Reduce the constrain of the model (e.g., reduce the regularization hyperparameter).

17. What is a test set, and why would you want to use it?
We split the data into two sets: The training set and the test set. We train the model using the training set and test it using the test set.
We evaluate the model(error rate on new cases) using the test set which tell us how the model will perfomr on instances it has never seen before.

18. What is the purpose of a validation set?
We use holdout validation to hold out part of the training set to evaluate several candidates models and select the best one. The new hold-out set is called the validation set.

Here, how we can use it:

  • Holdout Validation: Train multiple models with various hyperparameters on the reduced training set (the full training set minus the validation set).
  • Train the best model on the full training set(including the validation set), which gives the final model.
  • Evaluate this final model on the test set to get an estimate of the generalize error.

19. What is the train-dev set, when do you need it, and how do you use it?
The train-dev set is used when there is risk of mismatch between the training data and the data used in the validation and test datasets.
The train-dev set is a part of the training data set that’s held out. The model is trained on the rest of the training set, and evaluated on both the train-dev set and the validation set.

  • If the model performs well on the training set but not on the train-dev set, then the model is likely overfitting the training set.
  • If the model performs well on both the training set and the train-dev set, but not on the validation set, then there is probably a significant data mismatch between the training data and the validation + test data. And we should try to improve the trainig data to make it look like more like the validation + test data.
Let’s visualize the splitting (Github style) 
├──Data
├ ├──training set
├ ├ ├──training set
├ ├ ├ ├──training set
├ ├ ├ ├──train-dev set
├ ├ ├──validation set
├ ├──Test set

20. What can go wrong if you tune hyperparameters using the test set?
If you tune hyperparameters using the test set, you risk overfitting the test set.