BenJoyenConseil / algolang

Decision Tree, Random Forest implementations with golang thanks to Jason Brownlee's work

Algolang

Machine Learning algorithms : Decision Tree, Random Forest implementations

Algorithms & Usage

"This section applies the CART algorithm to the Bank Note dataset."

Load CSV file into master Matrix

types := map[string]string{"y": "float"}
df := io.LoadCsv("./testdata/data_banknote_authentication.txt", csv.Headers([]string{"col_0", "col_1", "col_2", "col_3", "y"}), csv.Types(types))
m := io.ToMatrix(df)

The cross validation with the accuracy score on DecisionTree

scores := eval.CrossVal(m, 4, 5, decision.Fit, map[string]int{"maxDepth": 5, "minSize": 10})
fmt.Println("Decision Tree", scores)

Output :

Decision Tree [95.25547445255475 98.17518248175182 96.71532846715328 90.51094890510949 98.91304347826086]

The cross validation with the accuracy score on RandomForest

scores = eval.CrossVal(m, 4, 5, ensemble.Fit, map[string]int{"n_estimator": 5, "maxDepth": 5, "minSize": 10})
fmt.Println("RandoForest", scores)

Output :

RandoForest [90.14598540145985 89.78102189781022 96.71532846715328 92.33576642335767 93.11594202898551]

Packages

• io / : it has ReadCSV that returns a QFrame (like pandas.DataFrame for golang). ToMatrix takes a QFrame and return a gonum.mat.Dense object

• mathelper / : matrix helpers like []float64 to gonum.mat.Vector convertion (into a Row or Column object). There is a Mode (statistic) function taking a gonum.mat.Vector

• eval / : has Accuracy score function in metric.go and expose CrossVal that takes an algo Fit function and return an array of the resultted accuracy scores for many folds

• algo /

  • model.go : defines the Model interface which has Predict contract.
  • decision / : DecisionTree is exposed by this package, using CART and the gini function.
  • ensemble / : RandomForest algorithm is exposed by this package. It uses Boostraping and Bagging of DecisionTrees.