Update notes

Jonas1312 · Sep 19, 2024 · 43ff391 · 43ff391
1 parent 8e5d554
commit 43ff391
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 4 deletions.
diff --git a/base/science-tech-maths/machine-learning/machine-learning.md b/base/science-tech-maths/machine-learning/machine-learning.md
@@ -45,7 +45,7 @@ A machine learning algorithm is an algorithm that is able to learn patterns from
   - [Generative models vs discriminative models](#generative-models-vs-discriminative-models)
   - [Ensemble methods](#ensemble-methods)
   - [Class imbalance](#class-imbalance)
-  - [Hyperparameter Optimization](#hyperparameter-optimization)
+  - [Hyperparameter Tuning and Optimization](#hyperparameter-tuning-and-optimization)
   - [Gradient descent](#gradient-descent)
     - [Momentum](#momentum)
     - [Adaptive learning rates](#adaptive-learning-rates)
@@ -524,7 +524,9 @@ Weighted loss functions vs weighted sampling?
 
 Many papers use the term long-tail learning to refer to class imbalance in multi-class classification tasks, so you can find lots of relevant research under this keyword
 
-## Hyperparameter Optimization
+## Hyperparameter Tuning and Optimization
+
+https://developers.google.com/machine-learning/guides/deep-learning-tuning-playbook
 
 - Babysitting: trial and error
 - Grid Search: exhaustive search over a grid of hyperparameters

diff --git a/base/science-tech-maths/machine-learning/metrics/metrics.md b/base/science-tech-maths/machine-learning/metrics/metrics.md
@@ -15,8 +15,8 @@
 
 ## Precision-Recall curve vs ROC curve
 
-- Precision: how often the classifier is correct when it predicts positive $PRE = \frac{TP}{TP+FP}$ "Of all the apples I picked from the basket, how many are actually good?"
-- Recall: how often the classifier is correct for all positive instances $REC = \frac{TP}{TP+FN}$ "Of all the good apples available, how many did I actually pick?"
+- Precision: $PRE = \frac{\text{Relevant retrieved instances}}{\text{All retrieved instances}}= \frac{TP}{TP+FP}$ "Of all predicted positive, how many are actually positive?"
+- Recall: $REC = \frac{\text{Relevant retrieved instances}}{\text{All relevant instances}} = \frac{TP}{TP+FN}$ "Of all real positive cases, how many did we predict as positive?"
 - [The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0118432)
 - Indeed, ROC is useful when evaluating general-purpose classification, while AUPRC is the superior method when classifying rare events.
 - <https://towardsdatascience.com/why-you-should-stop-using-the-roc-curve-a46a9adc728>