Karteikarten
Classification 2 -LH
Karten in diesem Stapel (8)
- 1Key question?Is there difference between the sample means large w.r.t the scatter of points within each sample?
Is |m1-m2|large w.r.t.s^2? - 2Slightly more quantitative:If samples 1 and 2 appear to be normally distributed, then the sample means (m1 and m2) are good estimates of the population means (u1 and u2).
- 3If the samples 1 and 2 appear to have approximately the same amount of scatter,then a good estimate of the overall sample variance is s squared.
- 4Criterion function - scoring each w vector to find the best separation between the two projected samples?
- 5If our samples, after projection to the line defined by optimal w, are normally distributed with equal scatter:Then the point along the line defined by optimal z that minimizes classification error sits midway between the projected means. The decision boundary intersects this point at right angles to the line defined by optimal w.
- 6Classification 2 uses.Prediction (statistical inference) and decoding
- 7Post-processing techniquesProbability density functions.
Take into consideration context (smoke alarm), misses are more costly. Therefore adjust criterion accordingly, allowing for more false alarms and fewer misses. - 8The design process is heuristic - what does this mean.heuristic - proceeding to a solution by trial and error. Double back and incorporate more domain knowledge in feature extraction methods.
Further: After evaluation (and visualization) you may revisit preceding stages: Did you collect enough data? Could domain-dependent knowledge be used to extract more features? Can another classifier (quadratic SVM) better parse the feature space?
Informationen
- Beschreibung
- Karteikarten am Classification 2 -LH, erstellt von Amber Turner am 09/11/2020.
- Erstellt von
- Amber Turner
- Aufrufe
- 0
- Bewertung