Basics

• AI

  • machine learning

    • reinforcement learning

      • complex mathematics

        • game theory

    • supervised learning

      • training data

        • target variable a.k.a. label

        • features in obeservation i.e. examples in training data

          • model

            • classification: assiging a category to an obsevation

              • outcome: category

            • regression: assign a continious variable

              • outcome: value

            • find conncetion from features and labels

              • predict label from features

    • unsupervised learning

      • training data

        • features in obeservation i.e. examples in training data

          • model

            • find patterns in features

              • cluster into group by features

                • find events that happen together as associations

                • find outliers as anormalities

    • deep learning

      • highly complex taks with millions of features

        • mimic the brain's neural network with descision nodes

          • labels for the obsevations are give, and the networks buids itself on the obsevations' features

Model Building

1. features

   • find

   • extract

2. split data set

   • train dataset

   • test dataset a.k.a. unseen data

3. train model

4. evaluate model

   • test dataset

     • supervised learning

       • overfitting: great with training data, poor on test data

         • model memorizes, but does not generalise

         • accuracy

           • correctly predicted per all predictions

           • correctly predicted per all all events that should have be predicted

         • confusion matrix

         • evaluating regression

           • root mean square error etc.

     • unsupervised learning

       • no prediction, not testing of prediction

   • accept or re-train model

     • retrain

       • dimensionality reduction

         • meaningless features

         • correlated similar features

       • hyperparameter tuning

       • ensemble methods

         • model combination