KnetMetrics

A standalone machine learning metrics library implemented in pure Julia by Emirhan Kurtuluş. A vast collection of classification, regression and pairwise metrics and related visualizations are included. This package is created as a part of the Knet ecosystem; however, built-in Julia arrays and all other types that support the same set of operations are compatible.

julia> Pkg.add("KnetMetrics"); using KnetMetrics
# dummy data
julia> y_true = [2, 3, 2, 2, 2, 1, 3, 3, 2, 1, 2, 3, 2, 3, 2, 3, 1,1];
julia> y_pred = [1, 3, 1, 1, 3, 2, 2, 1, 3, 2, 3, 3, 2, 3, 2, 1, 1, 3];
# creating a confusion matrix
julia> c = confusion_matrix( y_true, y_pred, labels = [1,2,3]) #labels are truly optional

            Expected

      1      2      3
_____________________
      1      2      1   │1
      3      2      3   │2
      2      1      3   │3      Predicted

# testing some metrics
julia> f1_score(c) # f1_score(y_true, y_pred)
0.32307692307692304

julia> f1_score(c, average = "binary")
3-element Array{Float64,1}:
 0.2
 0.3076923076923077
 0.4615384615384615

julia> f1_score(c, average = "binary", normalize=true)
3-element Array{Float64,1}:
 0.33918173268560714
 0.5218180502855494
 0.7827270754283241

julia> f1_score(c, class_name = 3)
0.4615384615384615

julia> matthews_correlation_coeff(c, average = "micro")
0.20396752553080869

julia> matthews_correlation_coeff(c, average = "weighted")
0.07138375997792953

julia> matthews_correlation_coeff(c, average = "sample-weights", weights = [3,2,1])
0.03263110671272045

julia> minkowski_distance(y_true, y_pred)
21

julia> mean_absolute_error(y_true, y_pred)
0.8333333333333334

Note: (*) symbol denotes that the function has a built-in visualization through visualize function.

• Confusion Matrix *
• Condition Positive and Negative *
• Predicted Positive and Negative *
• Correctly and Incorrectly Classified
• True Positive Rate (Sensitivity Score, Recall Score) *
• True Negative Rate (Specificity Score) *
• Positive Predictive Value (Precision Score) *
• Accuracy Score *
• Balanced Accuracy Score *
• Negative Predictive Value *
• False Negative Rate *
• False Positive Rate *
• False Discovery Rate *
• False Omission Rate *
• Fbeta Score (F1 Score) *
• Prevalence Threshold *
• Threat Score *
• Matthews Correlation Coefficient *
• Fowlkes Mallows Index *
• Informedness *
• Markedness *
• Cohen Kappa Score
• Hamming Loss
• Jaccard Score *

• Maximum Residual Error
• Mean Absolute Error
• Mean Squared Error
• Mean Squared Log Error
• Median Absolute Error
• Mean Absolute Percentage Error

• Minkowski Distance
• Euclidian Distance
• Manhattan Distance
• Chebyshev Distance
• Braycurtis Distance
• Canberra Distance
• Cityblock Distance
• Mahalanobis Distance
• Correlation
• Cosine Distance
• Cosine Similarity

1. A greater range of Roc Curve related functions
2. A greater range of regression functions
3. A greater range of pairwise functions and kernels
4. Clustering metrics