ゾンビでわかる分類評価指標

F0e950c6bf20815f3d169f063d3c5d4f?s=47 Aipa
March 09, 2019

 ゾンビでわかる分類評価指標

PythonBeginners沖縄(ゼロからDeep1&DjangoでWebアプリ)@琉球大学 で発表してきたLT

https://python-beginners-okinawa.connpass.com/event/121317/

F0e950c6bf20815f3d169f063d3c5d4f?s=128

Aipa

March 09, 2019
Tweet

Transcript

  1. 4.
  2. 26.
  3. 31.

    ࠞಉߦྻʢconfusion matrixʣ • ෼ྨ໰୊ͷ݁ՌͷੑೳΛ໌Β͔ʹ͢Δߦྻ • ਅཅੑ(true positive → TP) •

    ਅӄੑ(true negative → TN) • ِཅੑ(false positive → FP) • ِӄੑ(false negative → FN)
  4. 37.

    ʢ࠶ܝʣκϯϏ͕50ਓ(?) ੜଘऀ͕50໊͖ͨ ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ 

    51  '/ κϯϏ ͡Όͳ͍  '1  5/ • (27 + 45) / (27 + 45 + 5 + 23) = 72% • ਖ਼ղ཰͸72%
  5. 41.

    κϯϏ͕10ਓ(?) ੜଘऀ͕90໊͖ͨ ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ 

    51  '/ κϯϏ ͡Όͳ͍  '1  5/ • (0 + 90) / (0 + 90 + 0 + 10) = 90% • ਖ਼ղ཰͸90%
  6. 43.

    κϯϏ͕10ਓ(?) ੜଘऀ͕90໊͖ͨ ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ 

    51  '/ κϯϏ ͡Όͳ͍  '1  5/ • (0 + 90) / (0 + 90 + 0 + 10) = 90% • ਖ਼ղ཰͸90% ͓Θ͔Γ͍͚ͨͩͨͩΖ͏͔
  7. 44.
  8. 49.

    ద߹཰ʢPrecisionʣ • Ϟσϧ͕κϯϏͱ༧ଌͨ݁͠ՌʹɺͲΕ͚ͩ κϯϏؚ͕·Ε͍͔ͯͨ • TP / (TP + FP)

    • 0 / (0 + 0) = 0% ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ  51  '/ κϯϏ ͡Όͳ͍  '1  5/
  9. 51.

    ࠶ݱ཰ʢRecallʣ ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ  51

     '/ κϯϏ ͡Όͳ͍  '1  5/ • κϯϏʹର͠Ϟσϧ͕κϯϏͱ൑அׂͨ͠߹ • TP / (TP + FN) • 0 / (0 + 10) = 0%
  10. 53.

    F஋ʢF1-scoreʣ • PrecisionͱRecallΛ૊Έ߹Θͤͯௐ੔ͨ͠஋ • 2 * Precision * Recall /

    Precision + Recall • 2 * 0 * 0 / 0 + 0 = 0% ༧ଌ͞ΕͨΫϥε ࣮ࡍͷΫϥε κϯϏͱ༧ଌ κϯϏ ͡Όͳ͍ͱ ༧ଌ κϯϏ  51  '/ κϯϏ ͡Όͳ͍  '1  5/
  11. 55.