Demystifying Ml, DL and AI

Editor's Notes

• #7: Data Science Workflow Define the Problem What is the problem? Provide formal and informal definitions. Why does the problem need to be solved? Motivation, benefits, how it will be used. How would I solve the problem? Describe how the problem would be solved manually to flush domain knowledge. Prepare Data Data Selection. Availability, what is missing, what can be removed. Data Preprocessing. Organize selected data by formatting, cleaning and sampling. Data Transformation. Feature engineering using scaling, attribute decomposition and attribute aggregation. Data visualizations such as with histograms. Spot Check Algorithms Test harness with default values. Run family of algorithms across all the transformed and scaled versions of dataset. View comparisons with box plots. Improve Results (Tuning) Algorithm Tuning: discovering the best models in model parameter space. This may include hyper parameter optimizations with additional helper services. Ensemble Methods: where the predictions made by multiple models are combined. Feature Engineering: where the attribute decomposition and aggregation seen in data preparation is tested further. Present Results Context (Why): how the problem definition arose in the first place. Problem (Question): describe the problem as a question. Solution (Answer): describe the answer the the question in the previous step. Findings: Bulleted lists of discoveries you made along the way that interests the audience. May include discoveries in the data, methods that did or did not work or the model performance benefits you observed. Limitations: describe where the model does not work. Conclusions (Why+Question+Answer)
• #8: Data Selection what data is available, what data is missing and what data can be removed. Data Preprocessing: organize, clean and sample. Data Transformation: scaling, attribute decomposition and attribute aggregation.
• #9: This is a subset of the available data that you need to train your ML/DL models. What is the extent of the data, where is it located and is there anything missing to solve your problem. Usually, this process is a little more involved with Machine Learning due to the data set types used to train and save Machine Learning models. With Machine Learning, more is not better, usually.
• #10: Formatting: related to data formats and schemas. ETL tools are great for this step. Cleaning: cleaning data is the removal or fixing of missing data. Sampling: sometimes you can get a smaller representation of your data to improve training times.
• #11: Scaling: provide consistency with values between 0 and 1 with standard units of measure. Decomposition: feature separation. Hour and time is an example. Aggregation: counts for login instead of full time stamp is an example.
• #12: Test Harness: The goal of the test harness is to be able to quickly and consistently test algorithms against a fair representation of the problem being solved. Performance Measure: classification, regression or clustering. Cross Validation: use the entire data set to train your model. In short this is to separate your data into a number of chunks (folds) except one and the final test is done on that fold. Testing Algorithms: test with groups
• #14: Regression Regression is actually a loose term because its and algebraic process. Ordinary Least Squares Regression (OLSR) Linear Regression Logistic Regression Stepwise Regression Multivariate Adaptive Regression Splines (MARS) Locally Estimated Scatterplot Smoothing (LOESS) Instance Based Also called winner-take-all methods and memory-based learning. Focus is put on the representation of the stored instances and similarity measures used between instances. k-Nearest Neighbor (kNN) Learning Vector Quantization (LVQ) Self-Organizing Map (SOM) Locally Weighted Learning (LWL) Regularization Penalizes more complex algorithms. Ridge Regression Least Absolute Shrinkage and Selection Operator (LASSO) Elastic Net Least-Angle Regression (LARS) Decision Tree Often fast and accurate, used for both classification and regression. Classification and Regression Tree (CART) Iterative Dichotomiser 3 (ID3) C4.5 and C5.0 (different versions of a powerful approach) Chi-squared Automatic Interaction Detection (CHAID) Decision Stump M5 Conditional Decision Trees Bayesian Used in classification and regression. Naive Bayes Gaussian Naive Bayes Multinomial Naive Bayes Averaged One-Dependence Estimators (AODE) Bayesian Belief Network (BBN) Bayesian Network (BN) Clustering Algorithms Organizes data into groups. k-Means k-Medians Expectation Maximisation (EM) Hierarchical Clustering Association Rule Association rule learning methods extract rules that best explain observed relationships between variables in data. Paints relationships between large multi-dimensional data sets. Apriori algorithm Eclat algorithm Artificial Neural Networks (ANN), usually included with Deep Learning The most popular artificial neural network algorithms are: Perceptron Back-Propagation Hopfield Network Radial Basis Function Network (RBFN) Deep Learning Used in semi-supervised learning Deep Boltzmann Machine (DBM) Deep Belief Networks (DBN) Convolutional Neural Network (CNN) Stacked Auto-Encoders Dimensionality Reduction Used to visualize dimensional data or to simplify data which can then be used in a supervised learning method. Principal Component Analysis (PCA) Principal Component Regression (PCR) Partial Least Squares Regression (PLSR) Sammon Mapping Multidimensional Scaling (MDS) Projection Pursuit Linear Discriminant Analysis (LDA) Mixture Discriminant Analysis (MDA) Quadratic Discriminant Analysis (QDA) Flexible Discriminant Analysis (FDA) Ensemble Boosting Bootstrapped Aggregation (Bagging) AdaBoost Stacked Generalization (blending) Gradient Boosting Machines (GBM) Gradient Boosted Regression Trees (GBRT) Random Forest
• #18: “Deep Neural Nets” was first coined by Hinton and has been used since then with Deep Learning.