@@ -30,6 +30,8 @@ An orchestrated and repeatable pattern which systematically transforms and proce
30304. Training the model;
31315. Testing the model;
3232
33+ ------------------------------------------------------------------------------------------------------------------------
34+ 33351. Asking the Right Question
3436-----------------------------
3537a. Define scope (including data sources);
@@ -45,7 +47,9 @@ d. Define how solution is created;
4547 - Use the Machine Learning Workflow to process and transform Pima Indian data to create a predictin model. This model
4648 must predict whih people are likely to develop diabetes with 70% or greater accuracy.
4749
48- 2. Preparing data
50+ ---------------------------------------------------------------------------------------------------------------------------
51+ 52+ 2. Preparing data
4953 ---------------------
5054 a. Tidy Data
5155 - Tidy datasets are easy to manipulate, model and visualize,and have a specific structure:
@@ -64,6 +68,9 @@ Data Rule #2:
6468* Columns to eliminate - Not used, no values, duplicates;
6569* Correlated columns - Same information in different format, add little value, and cause algorithm to get confused;
6670* Modling Data - Adjusting data types, creating columns, if required;
71+ * Dealing with missing data -
72+ - Ignore it - Algorithms may fail;
73+ - Impute it - update to "reasonable" values - Most frequent, Mean, Median, Expert reasonable value;
6774
6875Data Rule #3:
6976----------------
@@ -73,7 +80,83 @@ Data Rule #4:
7380--------------
7481Track how to manipulate data;
7582
76- 3.
83+ -------------------------------------------------------------------------------------------------------------------------
84+ 85+ 3. Selecting the algorithm:
86+ ------------------------------
87+ Role of the Algorithm
88+ - fit the training set and predict on the read data;
89+ - (fit()) training data -> Algorithm -> model;
90+ - (predict()) real data -> Model -> result;
91+ 92+ Over 50 algorithms
93+ - algorithm selection
94+ *. Compare factors;
95+ *. Difference of opinions about which factors are important;
96+ *. Develop your own factors;
97+ 98+ Algorithm Decision Factors
99+ --------------------------
100+ i. Learning Type
101+ ii. Result
102+ iii. Complexity
103+ iv. Basic vs Enhanced
104+ 105+ i. Learning Type:
106+ "Use the Machine Learning Workflow to process and transform Pima Indian data to create a "prediction model". This model must
107+ predict which people are likely to develop diabetes with 70% or greater accuracy."
108+ 109+ -> Prediction Model => Supervised machine learning;
110+ Over 28 algorithms
111+ 112+ ii. Result
113+ a. Regression - constinuous vales;
114+ b. Classification - discrete values;
115+ 116+ "Use the Machine Learning Workflow to process and transform Pima Indian data to create a prediction model. This model must
117+ "predict which people are likely to develop diabetes" with 70% or greater accuracy."
118+ 119+ - Diabetes
120+ - Binary (True/False)
121+ - Algorithm must support classification - Binary classification;
122+ ** Over 20 algorithms;
123+ 124+ iii. Complexity
125+ - Keep it simple;
126+ - Eliminate ensemble algorithms - Container algorithm; Multiple child algorithm, boost performance, Can be difficult to debug;
127+ ** Over 14 algorithm;
128+ 129+ iv. Enhanced vs. Basic
130+ - Enhanced - variation of basic, performance improvements, additional functionality, more complex;
131+ - Basic - simpler, easier to understand;
132+ 133+ Candidate Algorithms
134+ --------------------
135+ a. Naive Bayes;
136+ b. Logistics Regression;
137+ c. Decision Tree;
138+ 139+ a. Naive Bayes - Based on likelihood and probability; every feature has same weight; requires smaller amount of data;
140+ b. Logistic Regression - Binary result, relation between features are weighted;
141+ c. Decision Tree - Binary tree, node contains decision, requires enough data to determine nodes and splits;
142+ 143+ Selected algorithm - Naive Bayes
144+ -------------------------------
145+ Simple - easy to understand;
146+ Fast - up to 100X faster;
147+ Stable to data changes;
148+ 149+ Overview
150+ ----------
151+ Lots of algorithms available
152+ 153+ Selected based on
154+ - Learning = Supervised
155+ - Result = Binary classification
156+ - Non-ensemble
157+ - Basic
158+ 159+ 77160
78161
79162
0 commit comments