Ma kēia ʻatikala, e kālailai mākou i nā helu kumu o ka hoʻololi nā hana hoʻihoʻi laina в hana hoʻololi logit inverse (i kapa ʻia he hana pane logistic). A laila, e hoʻohana i ka arsenal ʻano likelihood kiʻekiʻe, e like me ke kumu hoʻohālike logistic regression, loaʻa iā mākou ka hana poho Nalo Logistic, a ma nā huaʻōlelo ʻē aʻe, e wehewehe mākou i kahi hana e koho ʻia ai nā ʻāpana o ka vector paona i ke kumu hoʻohālikelike logistic. .
ʻatikala:
- E haʻi hou kākou i ka pilina laina ma waena o nā mea hoʻololi ʻelua
- E ʻike kākou i ka pono o ka hoʻololi nā hana hoʻihoʻi laina в hana pane logistic
- E hoʻokō kākou i ka hoʻololi a me ka hoʻopuka hana pane logistic
- E ho'āʻo kākou e hoʻomaopopo i ke kumu i hewa ʻole ai ke ʻano o nā huinahā liʻiliʻi ke koho ʻana i nā ʻāpana oihana Nalo Logistic
- Hoʻohana mākou ʻano likelihood kiʻekiʻe no ka hooholo ana nā hana koho parameter :
5.1. Hanana 1: hana Nalo Logistic no nā mea me ka papa inoa 0 и 1:
5.2. Hanana 2: hana Nalo Logistic no nā mea me ka papa inoa -1 и +1:
Hoʻopiha ʻia ka ʻatikala me nā hiʻohiʻona maʻalahi kahi e maʻalahi ai ka helu ʻana ma ka waha a i ʻole ma ka pepa; i kekahi mau hihia, pono paha ka helu helu. No laila e mākaukau :)
Kuhi ʻia kēia ʻatikala no nā ʻepekema data me kahi pae mua o ka ʻike i nā kumu o ka aʻo ʻana i ka mīkini.
E hāʻawi pū ka ʻatikala i ke code no ke kaha kiʻi kiʻi a me ka helu ʻana. Ua kākau ʻia nā code a pau ma ka ʻōlelo ʻthlelo 2.7. E wehewehe mua wau e pili ana i ka "mea hou" o ka mana i hoʻohana ʻia - ʻo ia kekahi o nā kūlana no ka lawe ʻana i ka papa kaulana mai Yandex ma kahi kahua hoʻonaʻauao pūnaewele kaulana like Coursera, a, e like me ka manaʻo o kekahi, ua hoʻomākaukau ʻia nā mea ma muli o kēia papa.
01. Ka hilinaʻi laina pololei
He mea kūpono ke nīnau i ka nīnau - he aha ka pili o ka laina laina a me ka hoʻihoʻi logistic me ia?
He maʻalahi! ʻO ka regression logistic kekahi o nā hiʻohiʻona i pili i ka linear classifier. Ma nā huaʻōlelo maʻalahi, ʻo ka hana a ka mea hoʻokaʻawale laina ʻo ia ke wānana i nā waiwai i koho ʻia mai nā mea hoʻololi (regressors) . Manaʻoʻia ka hilinaʻi ma waena o nāʻano a me nā kumu waiwai laina laina. No laila ka inoa o ka classifier - linear. No ka hoʻohālikelike ʻana, ua hoʻokumu ʻia ke kumu hoʻohālikelike logistic ma luna o ka manaʻo he pilina laina ma waena o nā ʻano. a me nā kumu waiwai . ʻO kēia ka pilina.
Aia ka la'ana mua ma ke keena, a he pololei ia, e pili ana i ka hilina'i rectilinear o ka nui e a'o ana. I ke kaʻina hana o ka hoʻomākaukau ʻana i ka ʻatikala, ua loaʻa iaʻu kahi hiʻohiʻona i hoʻonohonoho mua i nā poʻe he nui - ʻo ka hilinaʻi o kēia manawa ma ka volta. ("Aliʻi hoʻi hou ʻana", N. Draper, G. Smith). E nana kakou maanei.
E like me ʻO ke kānāwai o Ohm:
kahi - ka ikaika o kēia manawa, - uila, - kū'ē.
Inā ʻaʻole mākou i ʻike Kānāwai o Ohm, a laila hiki iā mākou ke ʻike i ka hilinaʻi ma ka hoʻololi ʻana a me ke ana ana , oiai e kokua ana paa. A laila e ʻike mākou i ka pakuhi hilinaʻi от hāʻawi i kahi laina pololei a ʻoi aʻe paha ma o ke kumu. 'Ōlelo mākou "ʻoi aʻe a emiʻole" no ka mea,ʻoiai he pololei maoli ka pilina, hiki i kā mākou mau ana ke loaʻa nā hewa liʻiliʻi, a no laila,ʻaʻole e hāʻule pololei nā kiko ma ka pakuhi ma ka laina, akā e hoʻopuehu waleʻia a puni.
Paʻi 1 "Ka hilinaʻi" от »
Kiʻi palapala kiʻi kiʻi
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import random
R = 13.75
x_line = np.arange(0,220,1)
y_line = []
for i in x_line:
y_line.append(i/R)
y_dot = []
for i in y_line:
y_dot.append(i+random.uniform(-0.9,0.9))
fig, axes = plt.subplots(figsize = (14,6), dpi = 80)
plt.plot(x_line,y_line,color = 'purple',lw = 3, label = 'I = U/R')
plt.scatter(x_line,y_dot,color = 'red', label = 'Actual results')
plt.xlabel('I', size = 16)
plt.ylabel('U', size = 16)
plt.legend(prop = {'size': 14})
plt.show()
02. ʻO ka pono e hoʻololi i ka hoʻohālikelike laina laina
E nānā kākou i kekahi laʻana. E noʻonoʻo kākou e hana ana mākou i kahi panakō a ʻo kā mākou hana ʻo ka hoʻoholo ʻana i ka hiki ke uku ʻia ka mea hōʻaiʻē ma muli o kekahi mau kumu. No ka hoʻomaʻamaʻa ʻana i ka hana, e noʻonoʻo mākou i ʻelua mau kumu: ka uku o ka mea hōʻaiʻē a me ka nui o ka uku hōʻaiʻē.
He kūlana koʻikoʻi ka hana, akā me kēia hiʻohiʻona hiki iā mākou ke hoʻomaopopo i ke kumu ʻaʻole lawa ka hoʻohana nā hana hoʻihoʻi laina, a e ʻike pū i nā hoʻololi e pono ai ke hana me ka hana.
E hoʻi kākou i ka laʻana. Ua hoʻomaopopo ʻia ʻo ke kiʻekiʻe o ka uku, ʻoi aku ka nui o ka mea ʻaiʻē e hiki ke hoʻokaʻawale i kēlā me kēia mahina e uku i ka hōʻaiʻē. I ka manawa like, no kekahi mau uku uku e lilo kēia pilina i laina laina. No ka laʻana, e lawe mākou i kahi uku uku mai 60.000 RUR a i 200.000 RUR a manaʻo ʻia i loko o ka palena uku i ʻōlelo ʻia, ʻo ka hilinaʻi o ka nui o ka uku mahina ma ka nui o ka uku uku. E ʻōlelo mākou no ka laulā o ka uku i hōʻike ʻia ʻaʻole hiki ke hāʻule ka lakio o ka uku i ka uku ma lalo o 3 a pono e loaʻa i ka mea hōʻaiʻē he 5.000 RUR i mālama ʻia. A i kēia hihia wale nō, e manaʻo mākou e uku ka mea hōʻaiʻē i ka hōʻaiʻē i ka panakō. A laila, e lawe ʻia ke ʻano hoʻohālikelike linear regression:
kahi , , , - uku -ʻo ka mea hōʻai'ē, - uku hōʻaiʻē -ka mea hōʻaiʻē.
Hoʻololi i ka uku uku a me ka uku hōʻaiʻē me nā ʻāpana paʻa i ka hoʻohālikelike Hiki iā ʻoe ke hoʻoholo inā e hoʻopuka a hōʻole paha i kahi hōʻaiʻē.
Ke nānā nei mākou i mua, ʻike mākou, me nā ʻāpana i hāʻawi ʻia hana hoʻihoʻi laina, hoʻohana ʻia ma nā hana pane logistic e hoʻopuka i nā waiwai nui e hoʻopili i ka helu ʻana e hoʻoholo ai i ka hiki ke uku ʻia ka hōʻaiʻē. No laila, manaʻo ʻia e hoʻemi i kā mākou coefficients, e ʻōlelo kākou, e 25.000 mau manawa. ʻAʻole e hoʻololi kēia hoʻololi i nā coefficient i ka hoʻoholo e hoʻopuka i kahi hōʻaiʻē. E hoʻomanaʻo kākou i kēia wahi no ka wā e hiki mai ana, akā i kēia manawa, i mea e maopopo ai ka mea a mākou e kamaʻilio nei, e noʻonoʻo kākou i ke kūlana me ʻekolu mau mea hōʻaiʻē.
Papa 1 "Nā poʻe hōʻaiʻē kūpono"
Code no ka hana ʻana i ka pākaukau
import pandas as pd
r = 25000.0
w_0 = -5000.0/r
w_1 = 1.0/r
w_2 = -3.0/r
data = {'The borrower':np.array(['Vasya', 'Fedya', 'Lesha']),
'Salary':np.array([120000,180000,210000]),
'Payment':np.array([3000,50000,70000])}
df = pd.DataFrame(data)
df['f(w,x)'] = w_0 + df['Salary']*w_1 + df['Payment']*w_2
decision = []
for i in df['f(w,x)']:
if i > 0:
dec = 'Approved'
decision.append(dec)
else:
dec = 'Refusal'
decision.append(dec)
df['Decision'] = decision
df[['The borrower', 'Salary', 'Payment', 'f(w,x)', 'Decision']]
E like me ka ʻikepili i ka papaʻaina, ʻo Vasya, me ka uku o 120.000 RUR, makemake e loaʻa kahi hōʻaiʻē i hiki iā ia ke uku i kēlā me kēia mahina ma 3.000 RUR. Ua hoʻoholo mākou no ka ʻae ʻana i ka hōʻaiʻē, ʻoi aku ka uku o Vasya ma mua o ʻekolu manawa o ka nui o ka uku, a he 5.000 RUR i koe. Hoʻokō ʻo Vasya i kēia koi: . ʻOiai he 106.000 RUR koe. ʻOiai ke helu ʻana ua hōʻemi mākou i nā pilikia 25.000 mau manawa, ua like ka hopena - hiki ke ʻae ʻia ka hōʻaiʻē. E loaʻa pū ʻo Fedya i kahi hōʻaiʻē, akā ʻo Lesha, ʻoiai ʻo ia ka mea i loaʻa iā ia ka hapa nui, pono e kāohi i kāna mau ʻai.
E kahakii kakou i ka pakuhi no keia hihia.
Mahele 2 "Ka helu ʻana o nā mea hōʻaiʻē"
Code no ke kaha kiʻi
salary = np.arange(60000,240000,20000)
payment = (-w_0-w_1*salary)/w_2
fig, axes = plt.subplots(figsize = (14,6), dpi = 80)
plt.plot(salary, payment, color = 'grey', lw = 2, label = '$f(w,x_i)=w_0 + w_1x_{i1} + w_2x_{i2}$')
plt.plot(df[df['Decision'] == 'Approved']['Salary'], df[df['Decision'] == 'Approved']['Payment'],
'o', color ='green', markersize = 12, label = 'Decision - Loan approved')
plt.plot(df[df['Decision'] == 'Refusal']['Salary'], df[df['Decision'] == 'Refusal']['Payment'],
's', color = 'red', markersize = 12, label = 'Decision - Loan refusal')
plt.xlabel('Salary', size = 16)
plt.ylabel('Payment', size = 16)
plt.legend(prop = {'size': 14})
plt.show()
No laila, kā mākou laina pololei, i kūkulu ʻia e like me ka hana , hoʻokaʻawale i nā mea hōʻaiʻē "ʻino" mai nā mea "maikaʻi". ʻO kēlā mau mea hōʻaiʻē ʻaʻole i kūlike ka makemake me ko lākou hiki ma luna o ka laina (Lesha), aʻo ka poʻe, e like me nā palena o kā mākou kumu hoʻohālike, hiki ke uku i ka hōʻaiʻē ma lalo o ka laina (Vasya a me Fedya). I nā huaʻōlelo ʻē aʻe, hiki iā mākou ke ʻōlelo i kēia: ʻo kā mākou laina pololei e hoʻokaʻawale i nā mea hōʻaiʻē i ʻelua papa. E hōʻike mākou iā lākou penei: i ka papa E hoʻokaʻawale mākou i kēlā poʻe ʻaiʻē e uku i ka hōʻaiʻē ai ole ia, E hoʻokomo mākou i kēlā poʻe ʻaiʻē i hiki ʻole ke uku i ka hōʻaiʻē.
E hōʻuluʻulu mākou i nā hopena mai kēia hiʻohiʻona maʻalahi. E lawe kāua i kahi manaʻo a me ka hoololi ana i na hookuonoono o ke kiko i ka hoohalike like o ka laina , e noʻonoʻo i ʻekolu mau koho:
- Inā aia ke kiko ma lalo o ka laina a hāʻawi mākou iā ia i ka papa , a laila ka waiwai o ka hana e maikaʻi mai i luna . ʻO ia ke ʻano hiki iā mākou ke manaʻo i ka hiki ke uku i ka hōʻaiʻē i loko . ʻO ka nui o ka waiwai hana, ʻoi aku ka kiʻekiʻe o ka probability.
- Inā he kiko ma luna o kahi laina a hāʻawi mākou iā ia i ka papa ai ole ia, , a laila, ʻino ka waiwai o ka hana mai i luna . A laila e manaʻo mākou aia i loko o ka hiki ke hoʻihoʻi ʻaiʻē a, ʻoi aku ka nui o ka waiwai paʻa o ka hana, ʻoi aku ka kiʻekiʻe o ko mākou hilinaʻi.
- Aia ke kiko ma ka laina pololei, ma ka palena ma waena o nā papa ʻelua. Ma keia hihia, ka waiwai o ka hana e like ana a ua like ka likelika o ka uku ana i ka aie .
I kēia manawa, e noʻonoʻo kākou ʻaʻole ʻelua mau kumu, akā he mau kakini, ʻaʻole ʻekolu, akā he mau tausani o nā mea hōʻaiʻē. A laila ma kahi o kahi laina pololei e loaʻa iā mākou m-dimensional mokulele a me nā coefficients ʻAʻole mākou e lawe ʻia mai ka lewa lahilahi, akā loaʻa e like me nā lula āpau, a ma ke kumu o ka ʻikepili i hōʻiliʻili ʻia ma nā mea hōʻaiʻē i loaʻa a ʻaʻole i uku i ka hōʻaiʻē. A ʻoiaʻiʻo, e hoʻomaopopo mākou ke koho nei mākou i nā mea hōʻaiʻē me ka hoʻohana ʻana i nā coefficient i ʻike mua ʻia . ʻO kaʻoiaʻiʻo, ʻo ka hana o ka loiloi logistic regression e hoʻoholo pololei i nā ʻāpana , i ka waiwai o ka hana poho Nalo Logistic e hele i ka liʻiliʻi loa. Akā e pili ana i ka helu ʻana o ka vector , e ʻike hou aku mākou ma ka ʻāpana 5 o ka ʻatikala. I kēia manawa, hoʻi mākou i ka ʻāina i hoʻohiki ʻia - i kā mākou panakō a me kāna mau mea kūʻai aku ʻekolu.
Mahalo i ka hana ʻike mākou i ka mea hiki ke hāʻawi ʻia i kahi hōʻaiʻē a pono e hōʻole ʻia. Akā ʻaʻole hiki iā ʻoe ke hele i ke alakaʻi me ia ʻike, no ka mea makemake lākou e kiʻi mai iā mākou i ka hiki ke uku ʻia ka hōʻaiʻē e kēlā me kēia mea ʻaiʻē. He aha ka hana? He maʻalahi ka pane - pono mākou e hoʻololi i ka hana , nona nā waiwai i loko o ka laulā i kahi hana nona nā waiwai e waiho ʻia ma ka laulā . A aia kekahi hana, ua kapa ʻia hana pane logistic a i ʻole hoʻololi hoʻololi-logit. Hui:
E ʻike kākou i kēlā me kēia ʻanuʻu pehea e hana ai hana pane logistic. E hoʻomaopopo e hele mākou ma ke ala ʻē aʻe, ʻo ia hoʻi. e manaʻo mākou ua ʻike mākou i ka waiwai probability, aia ma ka laulā mai i luna a laila mākou e "unwind" i kēia waiwai i ka laulā o nā helu mai i luna .
03. Loaʻa iā mākou ka hana pane logistic
KaʻAnuʻu Hana 1. E hoʻohuli i nā waiwai kūpono i loko o kahi laulā
I ka wā o ka hoʻololi ʻana o ka hana в hana pane logistic E haʻalele mākou i kā mākou loiloi hōʻaiʻē a hele i kahi mākaʻikaʻi o ka poʻe bookmaker ma kahi. ʻAʻole, ʻoiaʻiʻo, ʻaʻole mākou e kau i nā bets, ʻo nā mea a pau e makemake ai iā mākou aia ke ʻano o ka ʻōlelo, no ka laʻana, ʻo ka manawa kūpono ʻo 4 a 1. ʻO nā kuʻuna, ʻike ʻia e nā bettors āpau, ʻo ia ka ratio o nā "kūpono" i " hemahema”. Ma nā huaʻōlelo probability, ʻo ka ʻokoʻa ka likelika o kahi hanana i puʻunaue ʻia e ka likelika o ka hanana ʻaʻole i hiki. E kākau kākou i lalo i ke ʻano no ka loaʻa ʻana o kahi hanana :
kahi - hiki i kahi hanana i hiki mai, — hiki i kekahi hanana ʻAʻole i hiki mai
No ka laʻana, inā e hahau ka lio ʻōpio ikaika a pāʻani i kapa ʻia ʻo "Veterok" i kahi luahine ʻelemakule a paʻakikī i kapa ʻia ʻo "Matilda" ma ka heihei. , a laila ʻo ka manawa o ka holomua no "Veterok". к a ʻo ke ʻano hoʻi, ʻo ka ʻike ʻana i nā pilikia, ʻaʻole ia e paʻakikī iā mākou ke helu i ka probability :
No laila, ua aʻo mākou e "unuhi" i ka probability i nā manawa kūpono, e lawe i nā waiwai mai i luna . E hana hou kākou i hoʻokahi ʻanuʻu a aʻo i ka "unuhi" i ka hiki i ka laina helu holoʻokoʻa mai i luna .
KaʻAnuʻu Hana 2. E hoʻohuli i nā waiwai kūpono i loko o kahi laulā
He mea maʻalahi loa kēia ʻanuʻu - e lawe kākou i ka logarithm o nā ʻokoʻa i ke kumu o ka helu Euler. a loaʻa iā mākou:
I kēia manawa ua ʻike mākou inā , a laila e helu i ka waiwai e maʻalahi loa a, ʻoi aku ka maikaʻi: . He oiaio keia.
Ma muli o ka hoihoi, e nānā kākou pehea inā , a laila manaʻo mākou e ʻike i kahi waiwai maikaʻi ʻole . Nānā mākou: . Pololei kēnā.
I kēia manawa ua ʻike mākou pehea e hoʻololi ai i ka waiwai probability mai i luna ma ka laina helu holoʻokoʻa mai i luna . Ma ka ʻanuʻu aʻe e hana ʻē mākou.
I kēia manawa, ʻike mākou e like me nā lula o ka logarithm, ʻike i ka waiwai o ka hana , hiki iā ʻoe ke helu i nā pilikia:
ʻO kēia ʻano o ka hoʻoholo ʻana i nā mea e pono ai mākou i ka pae aʻe.
KaʻAnuʻu 3. E kiʻi kākou i ke kumu e hoʻoholo ai
No laila ua aʻo mākou, ʻike , huli i nā waiwai hana . Eia nō naʻe, ʻoiaʻiʻo, pono mākou i ka ʻokoʻa - ʻike i ka waiwai loaʻa . No ka hana ʻana i kēia, e huli mākou i kahi manaʻo e like me ka hana inverse odds, e like me ia:
I loko o ka ʻatikala ʻaʻole mākou e kiʻi i ke ʻano o luna, akā e nānā mākou me ka hoʻohana ʻana i nā helu mai ka laʻana ma luna. Ua ʻike mākou me ka nui o 4 a 1 (), ʻo 0.8 ka nui o ka hanana o ka hanana.). E hoʻololi kākou: . Kūlike kēia me kā mākou helu helu i hana mua ʻia. E neʻe kāua.
Ma ka ʻanuʻu hope loa mākou i unuhi i kēlā , 'o ia ho'i, hiki iā 'oe ke ho'ololi i ka hana inverse odds. Loaʻa iā mākou:
E puunaue i ka helu a me ka helu me , A laila:
ʻO ka hihia wale nō, e hōʻoia ʻaʻole mākou i hana hewa ma nā wahi āpau, e hana mākou i hoʻokahi kikoʻī liʻiliʻi. Ma ka ʻanuʻu 2, mākou no hoʻoholo i kēlā . A laila, hoʻololi i ka waiwai i loko o ka hana pane logistic, manaʻo mākou e loaʻa . Hoʻololi mākou a loaʻa:
Hoʻomaikaʻi, e ka makamaka heluhelu, ua loaʻa a hoʻāʻo mākou i ka hana pane logistic. E nānā kākou i ka pakuhi o ka hana.
Kakau 3 "Hana pane Logistic"
Code no ke kaha kiʻi
import math
def logit (f):
return 1/(1+math.exp(-f))
f = np.arange(-7,7,0.05)
p = []
for i in f:
p.append(logit(i))
fig, axes = plt.subplots(figsize = (14,6), dpi = 80)
plt.plot(f, p, color = 'grey', label = '$ 1 / (1+e^{-w^Tx_i})$')
plt.xlabel('$f(w,x_i) = w^Tx_i$', size = 16)
plt.ylabel('$p_{i+}$', size = 16)
plt.legend(prop = {'size': 14})
plt.show()
I loko o ka palapala hiki iā ʻoe ke ʻike i ka inoa o kēia hana hana sigmoid. Hōʻike maopopo ka pakuhi i ka hoʻololi nui ʻana i ka likelika o kahi mea pili i ka papa i loko o kahi ākea liʻiliʻi. , ma kahi mai i luna .
Manaʻo wau e hoʻi i kā mākou loiloi hōʻaiʻē a kōkua iā ia e helu i ka hiki ke uku ʻia ka hōʻaiʻē, inā ʻaʻole e waiho ʻia ʻo ia me ka ʻole o ka bonus :)
Papa 2 "Nā poʻe hōʻaiʻē kūpono"
Code no ka hana ʻana i ka pākaukau
proba = []
for i in df['f(w,x)']:
proba.append(round(logit(i),2))
df['Probability'] = proba
df[['The borrower', 'Salary', 'Payment', 'f(w,x)', 'Decision', 'Probability']]
No laila, ua hoʻoholo mākou i ke kūpono o ka uku hōʻaiʻē. Ma keʻano laulā, me he mea lā heʻoiaʻiʻo kēia.
ʻOiaʻiʻo, hiki iā Vasya, me ka uku o 120.000 RUR, hiki ke hāʻawi iā 3.000 RUR i ka panakō i kēlā me kēia mahina kokoke i 100%. Ma ke ala, pono mākou e hoʻomaopopo e hiki i kahi panakō ke hāʻawi i kahi hōʻaiʻē iā Lesha inā hāʻawi ke kulekele o ka panakō, no ka laʻana, no ka hōʻaiʻē ʻana i nā mea kūʻai aku me ka nui o ka uku hōʻaiʻē ma mua o, e ʻōlelo, 0.3. ʻO ia wale nō ma kēia hihia e hana ka panakō i kahi mālama nui no nā poho hiki ke hiki.
Pono e hoʻomaopopo ʻia ua lawe ʻia ka lakio o ka uku uku ma ka liʻiliʻi o 3 a me ka palena o 5.000 RUR mai ke kaupaku. No laila, ʻaʻole hiki iā mākou ke hoʻohana i ka vector o nā kaupaona ma kona ʻano kumu . Pono mākou e hōʻemi nui i nā coefficients, a ma kēia hihia mākou i puʻunaue i kēlā me kēia coefficient me 25.000, ʻo ia hoʻi, ma ke ʻano, hoʻoponopono mākou i ka hopena. Akā, ua hana ʻia kēia i mea e maʻalahi ai ka hoʻomaopopo ʻana i ka mea ma ka pae mua. I ke ola, ʻaʻole pono mākou e noʻonoʻo a hoʻoponopono i nā coefficients, akā loaʻa iā lākou. Ma nā ʻāpana aʻe o ka ʻatikala e loaʻa iā mākou nā hoohalike i koho ʻia ai nā ʻāpana .
04. ʻO ke ʻano huinahā liʻiliʻi no ka hoʻoholo ʻana i ka vector o nā kaupaona i ka hana pane logistic
Ua ʻike mua mākou i kēia ʻano no ke koho ʻana i kahi vector o nā kaupaona , ma ʻano huinahā liʻiliʻi loa (LSM) a i ka ʻoiaʻiʻo, no ke aha mākou e hoʻohana ʻole ai i nā pilikia hoʻohālikelike binary? ʻOiaʻiʻo, ʻaʻohe mea e pale iā ʻoe mai ka hoʻohana ʻana MNC, ʻO kēia ala wale nō i ka hoʻokaʻawale ʻana i nā pilikia e hāʻawi i nā hopena i ʻoi aku ka pololei Nalo Logistic. Aia ke kumu kumu no keia. E nānā mua kākou i kekahi laʻana maʻalahi.
E noʻonoʻo i kā mākou mau hiʻohiʻona (hoʻohana MSE и Nalo Logistic) ua hoʻomaka e koho i ka vector o nā kaupaona a ua hooki mākou i ka helu ʻana ma kekahi ʻanuʻu. ʻAʻole ia he mea nui inā i ka waena, i ka hopena a i ka hoʻomaka paha, ʻo ka mea nui ua loaʻa iā mākou kekahi mau waiwai o ka vector o nā kaupaona a e manaʻo mākou i kēia kaʻina, ka vector o nā kaupaona. no nā hiʻohiʻona ʻelua ʻaʻohe ʻokoʻa. A laila e lawe i nā paona hopena a hoʻololi iā lākou i loko hana pane logistic () no kekahi mea pili i ka papa . Ke nānā nei mākou i ʻelua mau hihia inā, e like me ka vector o nā kaupaona i koho ʻia, kuhi hewa loa kā mākou kumu hoʻohālike a me ka hope - ua hilinaʻi nui ke kumu hoʻohālike no ka papa ka mea. . E ʻike kākou i nā uku e hoʻopuka ʻia ke hoʻohana MNC и Nalo Logistic.
Code e helu i nā hoʻopaʻi ma muli o ka hana pohō i hoʻohana ʻia
# класс объекта
y = 1
# вероятность отнесения объекта к классу в соответствии с параметрами w
proba_1 = 0.01
MSE_1 = (y - proba_1)**2
print 'Штраф MSE при грубой ошибке =', MSE_1
# напишем функцию для вычисления f(w,x) при известной вероятности отнесения объекта к классу +1 (f(w,x)=ln(odds+))
def f_w_x(proba):
return math.log(proba/(1-proba))
LogLoss_1 = math.log(1+math.exp(-y*f_w_x(proba_1)))
print 'Штраф Log Loss при грубой ошибке =', LogLoss_1
proba_2 = 0.99
MSE_2 = (y - proba_2)**2
LogLoss_2 = math.log(1+math.exp(-y*f_w_x(proba_2)))
print '**************************************************************'
print 'Штраф MSE при сильной уверенности =', MSE_2
print 'Штраф Log Loss при сильной уверенности =', LogLoss_2
He hihia o ka hewa - hāʻawi ke kumu hoʻohālike i kahi mea i kahi papa me ka hikiwawe o 0,01
Hoʻopaʻi ma ka hoʻohana MNC e:
Hoʻopaʻi ma ka hoʻohana Nalo Logistic e:
He hihia hilinaʻi ikaika - hāʻawi ke kumu hoʻohālike i kahi mea i kahi papa me ka hikiwawe o 0,99
Hoʻopaʻi ma ka hoʻohana MNC e:
Hoʻopaʻi ma ka hoʻohana Nalo Logistic e:
Hōʻike maikaʻi kēia hiʻohiʻona inā he hewa nui ka hana poho Nalo Log hoʻopaʻi nui i ke kumu hoʻohālike ma mua o MSE. E hoʻomaopopo kākou i ke ʻano o ka manaʻo o ka hoʻohana ʻana i ka hana poho Nalo Log i nā pilikia hoʻonohonoho.
05. ʻO ke ʻano likelihood kiʻekiʻe a me ka hoʻihoʻi loiloi
E like me ka mea i ʻōlelo mua ʻia, ua piha ka ʻatikala me nā hiʻohiʻona maʻalahi. Aia i loko o ke keena kekahi hiʻohiʻona a me nā malihini kahiko - nā mea hōʻaiʻē panakō: Vasya, Fedya a me Lesha.
I ka hihia wale nō, ma mua o ka hoʻomohala ʻana i ka laʻana, e hoʻomanaʻo wau iā ʻoe i ke ola e pili ana mākou i kahi laʻana aʻo o nā tausani a i ʻole miliona mau mea me nā ʻumi a i ʻole haneli mau hiʻohiʻona. Eia naʻe, eia nā helu i lawe ʻia i hiki iā lākou ke komo maʻalahi i ke poʻo o kahi ʻepekema data novice.
E hoʻi kākou i ka laʻana. E noʻonoʻo kākou ua hoʻoholo ka luna o ka panakō e hāʻawi i kahi hōʻaiʻē i nā mea āpau e pono ai, ʻoiai ʻo ka algorithm i haʻi aku iā ia ʻaʻole e hoʻopuka iā Lesha. A i kēia manawa ua hala ka manawa a ʻike mākou ʻo wai o nā koa ʻekolu i uku i ka hōʻaiʻē a ʻo wai ʻaʻole. ʻO ka mea e manaʻo ʻia: Ua uku ʻo Vasya lāua ʻo Fedya i ka hōʻaiʻē, akā ʻaʻole ʻo Lesha. I kēia manawa, e noʻonoʻo kākou e lilo kēia hopena i mea hoʻomaʻamaʻa hou no mākou a, i ka manawa like, me he mea lā ua nalowale nā ʻikepili āpau i nā kumu e hoʻopiʻi i ka likelihood o ka uku ʻana i ka hōʻaiʻē (ka uku o ka mea hōʻaiʻē, ka nui o ka uku mahina). A laila, intuitively, hiki iā mākou ke manaʻo ʻaʻole ʻo kēlā me kēia kolu o ka mea hōʻaiʻē e uku i ka hōʻaiʻē i ka panakō, a i ʻole, ʻo ke ʻano o ka mea hōʻaiʻē hou e uku i ka hōʻaiʻē. . ʻO kēia manaʻo intuitive he hōʻoia theoretical a ua hoʻokumu ʻia ma ʻano likelihood kiʻekiʻe, ua kapa pinepine ia ma ka palapala kumukānāwai likelihood kiʻekiʻe.
ʻO ka mea mua, e kamaʻāina kākou i ka mea hana manaʻo.
Hiki i ka laʻana ʻo ia ke kūpono o ka loaʻa ʻana o kahi laʻana like, loaʻa pololei nā ʻike / hopena, ʻo ia hoʻi. ka huahana o nā probabilities o ka loaʻa ʻana o kēlā me kēia o nā hopena hoʻohālike (no ka laʻana, inā i uku ʻia ka hōʻaiʻē o Vasya, Fedya a me Lesha i ka manawa like).
Hana likelihood pili i ka likelihood o kahi laʻana i nā waiwai o nā ʻāpana mahele.
I kā mākou hihia, ʻo ka laʻana hoʻomaʻamaʻa he papahana Bernoulli maʻamau, kahi e lawe ai ka mea hoʻololi random i ʻelua mau waiwai: ai ole ia, . No laila, hiki ke kākau ʻia ka laʻana likelihood ma ke ʻano he hana likelihood o ka parameter penei:
Hiki ke unuhi ʻia ka mea kākau ma luna aʻe penei. ʻO ka like like e uku ai ʻo Vasya lāua ʻo Fedya i ka hōʻaiʻē , ua like ka like ole o Lesha e uku i ka aie (no ka mea, ʻaʻole ia ka uku hōʻaiʻē i hana ʻia), no laila ua like ka like like o nā hanana ʻekolu. .
ʻO ke ʻano likelihood kiʻekiʻe he ala no ka koho ʻana i kahi ʻāpana ʻike ʻole ma ka hoʻonui ʻana nā hana likelihood. I kā mākou hihia, pono mākou e ʻimi i kahi waiwai i ka manawa hea hiki i kona kiekie.
No hea mai ka manaʻo maoli - e ʻimi i ka waiwai o kahi ʻāpana ʻike ʻole kahi i hiki ai ka hana likelihood i kahi kiʻekiʻe? ʻO ke kumu o ka manaʻo ma muli o ka manaʻo he laʻana wale nō ke kumu o ka ʻike i loaʻa iā mākou e pili ana i ka heluna kanaka. Hōʻike ʻia nā mea a pau a mākou e ʻike ai e pili ana i ka heluna kanaka ma ka hāpana. No laila, ʻo kā mākou mea e ʻōlelo ai, ʻo kahi laʻana ka hōʻike pololei loa o ka heluna kanaka i loaʻa iā mākou. No laila, pono mākou e ʻimi i kahi ʻāpana kahi e lilo ai ka laʻana i loaʻa i mea kūpono loa.
ʻIke loa, ke hana nei mākou i kahi pilikia optimization kahi e pono ai mākou e ʻimi i ka palena kiʻekiʻe o kahi hana. No ka loaʻa ʻana o ke kiko extremum, pono e noʻonoʻo i ke kūlana o ka papa mua, ʻo ia hoʻi, e hoʻohālikelike i ka derivative o ka hana me ka ʻole a hoʻoholo i ka hoohalike e pili ana i ka palena i makemake ʻia. Eia nō naʻe, ʻo ka ʻimi ʻana i ka derivative o kahi huahana o nā mea he nui he hana lōʻihi; e pale aku i kēia, aia kahi ʻenehana kūikawā - ke hoʻololi i ka logarithm nā hana likelihood. No ke aha e hiki ai ia hoʻololi? E hoʻolohe kākou i ka ʻoiaʻiʻo ʻaʻole mākou e ʻimi nei i ka extremum o ka hana ponoʻī, a me ke kiko extremum, oia hoi ka waiwai o ka palena i ike ole ia i ka manawa hea hiki i kona kiekie. I ka neʻe ʻana i kahi logarithm, ʻaʻole e loli ka helu extremum (ʻoiai e ʻokoʻa ka extremum ponoʻī), no ka mea, he hana monotononic ka logarithm.
E like me ka mea i luna, e hoʻomau i ka hoʻomohala ʻana i kā mākou hiʻohiʻona me nā hōʻaiʻē mai Vasya, Fedya a me Lesha. E neʻe mua kākou i ka logarithm o ka hana likelihood:
I kēia manawa hiki iā mākou ke hoʻokaʻawale i ka ʻōlelo ma :
A ʻo ka hope, e noʻonoʻo i ke kūlana o ka papa mua - hoʻohālikelike mākou i ka derivative o ka hana me ka ʻole:
No laila, ʻo kā mākou manaʻo intuitive o ka hiki ke uku ʻia ka hōʻaiʻē ua aponoia ma ke ano.
Nui, akā he aha kā mākou e hana ai me kēia ʻike i kēia manawa? Inā mākou e manaʻo ʻaʻole e hoʻihoʻi kēlā me kēia kolu o ka mea ʻaiʻē i ke kālā i ka panakō, a laila e lilo ka hope i ka panakalupa. Pololei, akā i ka wā e loiloi ana i ka likelika o ka uku hōʻaiʻē e like me ʻAʻole mākou i noʻonoʻo i nā kumu e hoʻopiʻi ai i ka hōʻaiʻē: ka uku o ka mea hōʻaiʻē a me ka nui o ka uku mahina. E hoʻomanaʻo mākou ua helu mua mākou i ka hiki ke uku i ka hōʻaiʻē e kēlā me kēia mea kūʻai aku, me ka noʻonoʻo ʻana i kēia mau mea like. Maikaʻi ka loaʻa ʻana o nā probabilities ʻokoʻa mai ka like mau .
E wehewehe i ka likelihood o nā laʻana:
Code no ka helu ʻana i nā laʻana likelihood
from functools import reduce
def likelihood(y,p):
line_true_proba = []
for i in range(len(y)):
ltp_i = p[i]**y[i]*(1-p[i])**(1-y[i])
line_true_proba.append(ltp_i)
likelihood = []
return reduce(lambda a, b: a*b, line_true_proba)
y = [1.0,1.0,0.0]
p_log_response = df['Probability']
const = 2.0/3.0
p_const = [const, const, const]
print 'Правдоподобие выборки при константном значении p=2/3:', round(likelihood(y,p_const),3)
print '****************************************************************************************************'
print 'Правдоподобие выборки при расчетном значении p:', round(likelihood(y,p_log_response),3)
Laʻana likelihood i ka waiwai mau :
Laʻana likelihood i ka helu ʻana i ka hiki ke uku ʻia ka hōʻaiʻē me ka noʻonoʻo ʻana i nā kumu :
ʻO ka likelihood o kahi laʻana me ka probability i helu ʻia ma muli o nā kumu i ʻoi aku ka kiʻekiʻe ma mua o ka likelihood me ka waiwai probability mau. He aha ke ʻano o kēia? Hōʻike kēia i ka ʻike e pili ana i nā kumu i hiki ai ke koho pololei i ke kūpono o ka uku hōʻaiʻē no kēlā me kēia mea kūʻai. No laila, i ka wā e hoʻopuka ai i ka hōʻaiʻē aʻe, ʻoi aku ka pololei o ka hoʻohana ʻana i ke kumu hoʻohālike i manaʻo ʻia ma ka hopena o ka pauku 3 o ka ʻatikala no ka loiloi ʻana i ka hiki ke uku ʻia ka ʻaiʻē.
Akā, inā makemake mākou e hoʻonui hana likelihood laʻana, No ke aha e hoʻohana ʻole ai i kekahi algorithm e hoʻopuka i nā probabilities no Vasya, Fedya a me Lesha, no ka laʻana, e like me 0.99, 0.99 a me 0.01. Malia paha e hana maikaʻi ana kēlā algorithm i ka laʻana aʻo, no ka mea e hoʻokokoke mai ai ka waiwai likelihood i ka hāpana , akā, ʻo ka mea mua, ʻo ia algorithm e loaʻa i nā pilikia me ka hiki ke hoʻonui, a ʻo ka lua, ʻaʻole pololei kēia algorithm. A inā ʻaʻole i hoʻokomo ʻia nā ʻano o ka hakakā ʻana i ka overtraining (like me ka nāwaliwali generalization hiki) i ka hoʻolālā o kēia ʻatikala, a laila e hele kākou ma ka helu ʻelua i nā kikoʻī hou aku. No ka hana ʻana i kēia, e pane wale i kahi nīnau maʻalahi. Hiki i ka likelika o Vasya a me Fedya ke hoʻihoʻi i ka hōʻaiʻē, e noʻonoʻo i nā kumu i ʻike ʻia e mākou? Mai ka manaʻo o ka loiloi kani, ʻoiaʻiʻo ʻaʻole, ʻaʻole hiki. No laila e uku ʻo Vasya i ka 2.5% o kāna uku i kēlā me kēia mahina e uku i ka hōʻaiʻē, a ʻo Fedya - kokoke i 27,8%. Eia kekahi ma ka pakuhi 2 "Ka hoʻokaʻawale ʻana o nā mea kūʻai aku" ʻike mākou ua ʻoi aku ka nui o Vasya mai ka laina e hoʻokaʻawale ana i nā papa ma mua o Fedya. A ʻo ka hope, ʻike mākou i ka hana no Vasya a me Fedya i nā waiwai like ʻole: 4.24 no Vasya a me 1.0 no Fedya. I kēia manawa, inā loaʻa iā Fedya kahi kauoha o ka nui a i ʻole e noi i kahi hōʻaiʻē liʻiliʻi, a laila e like nā probabilities o ka uku ʻana i ka hōʻaiʻē no Vasya a me Fedya. ʻO ia hoʻi, ʻaʻole hiki ke hoʻopunipuni i ka hilinaʻi laina. A inā mākou i helu maoli i nā pilikia , a ʻaʻole i lawe iā lākou i waho o ka ea lahilahi, hiki iā mākou ke ʻōlelo palekana i kā mākou mau waiwai ʻae maikaʻi iā mākou e kuhi i ka hiki ke uku ʻia ka hōʻaiʻē e kēlā me kēia mea ʻaiʻē, akā no ka mea ua ʻae mākou e manaʻo i ka hoʻoholo ʻana o nā coefficients. ua hana ʻia e like me nā lula āpau, a laila e manaʻo mākou pēlā - ʻae kā mākou coefficients iā mākou e hāʻawi i kahi kuhi maikaʻi aʻe o ka probability :)
Eia naʻe, hoʻololi mākou. Ma kēia ʻāpana pono mākou e hoʻomaopopo i ke ʻano o ka hoʻoholo ʻana i ka vector o nā kaupaona , he mea pono e loiloi i ka hiki ke hoʻihoʻi ʻia ka hōʻaiʻē e kēlā me kēia mea ʻaiʻē.
E hōʻuluʻulu pōkole mākou me ka arsenal a mākou e hele ai e ʻimi i nā pilikia :
1. Manaʻo mākou he laina laina ka pilina ma waena o ka mea hoʻololi i manaʻo ʻia (waiwai wānana) a me ka mea e hoʻohuli ai i ka hopena. No kēia kumu hoʻohana ʻia hana hoʻihoʻi laina ʻoluʻolu , ka laina e hoʻokaʻawale i nā mea (nā mea kūʻai aku) i nā papa и ai ole ia, (nā mea kūʻai i hiki ke uku i ka hōʻaiʻē a me ka poʻe ʻaʻole). I kā mākou hihia, aia ke ʻano o ka hoohalike .
2. Hoʻohana mākou hana logit inverse ʻoluʻolu e hoʻoholo i ke kūpono o kekahi mea no ka papa .
3. Manaʻo mākou i kā mākou hoʻonohonoho hoʻomaʻamaʻa ʻana ma ke ʻano he hoʻokō o kahi laulā Bernoulli papahana, ʻo ia hoʻi, no kēlā me kēia mea, ua hoʻokumu ʻia kahi ʻano hoʻololi like ʻole, me ka likelika (ʻo ia no kēlā me kēia mea) lawe i ka waiwai 1 a me ka mea hiki - 0.
4. ʻIke mākou i ka mea e pono ai mākou e hoʻonui hana likelihood laʻana e noʻonoʻo ana i nā kumu i ʻae ʻia i lilo ka laʻana i loaʻa i mea kūpono loa. I nā huaʻōlelo ʻē aʻe, pono mākou e koho i nā ʻāpana kahi e kūpono ai ka laʻana. I kā mākou hihia, ʻo ka ʻāpana i koho ʻia ʻo ia ka hiki ke uku ʻia ka hōʻaiʻē , e pili ana i nā coefficient ʻike ʻole ʻia . No laila pono mākou e ʻimi i kahi vector o nā kaupaona , ma kahi e hiki ai ka nui o ka hāpana.
5. ʻIke mākou i ka mea e hoʻonui ai laʻana likelihood hana hiki iāʻoe ke hoʻohana ʻano likelihood kiʻekiʻe. A ʻike mākou i nā hana hoʻopunipuni āpau e hana me kēia ʻano.
ʻO kēia ke ʻano o ka neʻe ʻana i nā neʻe he nui :)
I kēia manawa e hoʻomanaʻo i ka hoʻomaka ʻana o ka ʻatikala makemake mākou e kiʻi i ʻelua ʻano hana poho Nalo Logistic ma muli o ke ʻano i koho ʻia ai nā papa mea. Ua hiki mai i ka hoʻokaʻawale ʻana i nā pilikia me nā papa ʻelua, ua kapa ʻia nā papa e like me и ai ole ia, . Ma muli o ka notation, e loaʻa i ka pukana kahi hana poho.
Hana 1. Ka hoʻokaʻawale ʻana o nā mea i loko и
Ma mua, i ka wā e hoʻoholo ai i ka likelihood o kahi laʻana, kahi i helu ʻia ai ka probability o ka uku ʻaiʻē e ka mea ʻaiʻē ma muli o nā kumu a hāʻawi ʻia i nā coefficients. , ua hoʻohana mākou i ke ʻano:
ʻOiaʻiʻo ʻo ia ke ʻano nā hana pane logistic no kekahi vector o nā kaupaona
A laila ʻaʻohe mea e pale iā mākou mai ke kākau ʻana i ka hana likelihood like penei:
Hiki i kekahi manawa he mea paʻakikī i kekahi poʻe loiloi novice e hoʻomaopopo koke i ke ʻano o kēia hana. E nānā kākou i 4 mau hiʻohiʻona pōkole e hoʻomaʻemaʻe i nā mea:
1. ina (ʻo ia hoʻi, e like me ka laʻana aʻo, no ka papa +1 ka mea), a me kā mākou algorithm hoʻoholo i ka likelika o ka hoʻokaʻawale ʻana i kekahi mea i ka papa e like me 0.9, a laila e helu ʻia kēia ʻāpana likelihood penei:
2. ina a me ka , a laila e like ka helu ʻana penei:
3. ina a me ka , a laila e like ka helu ʻana penei:
4. ina a me ka , a laila e like ka helu ʻana penei:
ʻIke ʻia e hoʻonui ʻia ka hana likelihood i nā hihia 1 a me 3 a i ʻole ma ka hihia maʻamau - me nā helu kuhi pololei o nā probabilities o ka hāʻawi ʻana i kahi mea i kahi papa. .
Ma muli o ka ʻoiaʻiʻo i ka wā e hoʻoholo ai i ka hiki ke hāʻawi i kahi mea i kahi papa ʻAʻole mākou ʻike wale i nā coefficients , a laila e ʻimi mākou iā lākou. E like me ka mea i ʻōlelo ʻia ma luna, he pilikia optimization kēia kahi e pono ai mākou e ʻimi mua i ka derivative o ka hana likelihood e pili ana i ka vector o nā kaupaona. . Eia naʻe, ʻo ka mea mua, he mea maʻalahi ke hoʻomaʻamaʻa i ka hana no mākou iho: e ʻimi mākou i ka derivative o ka logarithm. nā hana likelihood.
No ke aha ma hope o ka logarithm, in nā hana hewa logistic, ua hoʻololi mākou i ka hōʻailona mai maluna o . He mea maʻalahi nā mea a pau, no ka mea, i nā pilikia o ka loiloi ʻana i ka maikaʻi o kahi kumu hoʻohālike he mea maʻamau ka hōʻemi ʻana i ka waiwai o kahi hana, ua hoʻonui mākou i ka ʻaoʻao ʻākau o ka ʻōlelo me a no laila, ma kahi o ka hoʻonui ʻana, i kēia manawa mākou e hōʻemi i ka hana.
ʻOiaʻiʻo, i kēia manawa, i mua o kou mau maka, ua loaʻa pinepine ka hana poho - Nalo Logistic no kahi hoʻonohonoho hoʻomaʻamaʻa me nā papa ʻelua: и .
I kēia manawa, no ka loaʻa ʻana o nā coefficients, pono mākou e ʻimi i ka derivative nā hana hewa logistic a laila, me ka hoʻohana ʻana i nā ʻano loiloi helu, e like me ka gradient descent a i ʻole stochastic gradient descent, koho i nā coefficient maikaʻi loa. . Akā, ma muli o ka nui o ka ʻatikala, manaʻo ʻia e hoʻokō i ka hoʻokaʻawale ʻana iā ʻoe iho, a i ʻole he kumuhana kēia no ka ʻatikala aʻe me ka helu helu me ka ʻole o nā hiʻohiʻona kikoʻī.
Hana 2. Ka hoʻokaʻawale ʻana o nā mea i loko и
E like ka hoʻokokoke ʻana ma ʻaneʻi me nā papa и , akā ʻo ke ala ponoʻī i ka hoʻopuka o ka hana poho Nalo Logistic, e oi aku ka nani. E hoʻomaka kākou. No ka hana likelihood e hoʻohana mākou i ka mea hoʻohana "inā ... a laila..."... ʻO ia, inā No ka papa ka mea th , a laila e helu i ka likelihood o ka hāpana hoʻohana mākou i ka probability , inā pili ka mea i ka papa , a laila hoʻololi mākou i ka likelihood . Penei ke ano o ka hana likelihood:
E wehewehe mākou ma ko mākou mau manamana lima pehea e hana ai. E noʻonoʻo kākou i nā hihia 4:
1. ina и , a laila e "hele" ka la'ana.
2. ina и , a laila e "hele" ka la'ana.
3. ina и , a laila e "hele" ka la'ana.
4. ina и , a laila e "hele" ka la'ana.
ʻIke ʻia ma nā hihia 1 a me 3, i ka wā i hoʻoholo pololei ʻia nā probabilities e ka algorithm, hana likelihood e hoʻonui ʻia, ʻo ia hoʻi, ʻo ia ka mea a mākou i makemake ai e loaʻa. Eia naʻe, paʻakikī loa kēia ala a ma hope e noʻonoʻo mākou i kahi notation paʻa. Akā ʻo ka mua, e hoʻokaʻawale kākou i ka hana likelihood me ka hoʻololi ʻana i ka hōʻailona, no ka mea, i kēia manawa e hoʻemi mākou.
E pani kākou hoike :
E hoʻomāmā kākou i ka huaʻōlelo kūpono ma lalo o ka logarithm me ka hoʻohana ʻana i nā ʻenehana helu maʻalahi a loaʻa:
ʻO ka manawa kēia e hoʻopau ai i ka mea hoʻohana "inā ... a laila...". E hoʻomaopopo i ka wā o kahi mea no ka papa , a laila ma ka ʻōlelo ma lalo o ka logarithm, ma ka denominator, hookiekieia i ka mana , inā pili ka mea i ka papa , a laila hoʻokiʻekiʻe ʻia $e$ i ka mana . No laila, hiki ke maʻalahi ka notation no ka degere ma ka hoʻohui ʻana i nā hihia ʻelua i hoʻokahi: . ^ E Ha yM. A laila hana hewa logistic e lawe i ke ʻano:
E like me nā lula o ka logarithm, hoʻohuli mākou i ka hakina a hoʻopau i ka hōʻailona ""(minus) no ka logarithm, loaʻa iā mākou:
Eia ka hana poho poho logistic, i hoʻohana ʻia i ka hoʻonohonoho hoʻomaʻamaʻa me nā mea i hāʻawi ʻia i nā papa: и .
ʻAe, i kēia manawa ke haʻalele nei au a hoʻopau mākou i ka ʻatikala.
Nā mea kōkua
1. Palapala palapala
1) Hoʻopili i ka loiloi regression / N. Draper, G. Smith - 2nd ed. – M.: Waiwai a me Heluhelu, 1986 (unuhi mai ka olelo Pelekania)
2) Ka manaʻo kuhikahi a me ka helu makemakika / V.E. Gmurman - 9th ed. - M.: Kula Kiekie, 2003
3) Ka manaʻo kūpono / N.I. Chernova - Novosibirsk: Novosibirsk State University, 2007
4) Nāʻikepili pāʻoihana: mai kaʻikepili i kaʻike / Paklin N. B., Oreshkov V. I. - 2nd ed. — St. Petersburg: Peter, 2013
5) ʻEpekema ʻIkepili ʻepekema ʻikepili mai ka ʻōpala / Joel Gras - St. Petersburg: BHV Petersburg, 2017
6) Heluhelu kūpono no nā loea ʻIke ʻIkepili / P. Bruce, E. Bruce - St. Petersburg: BHV Petersburg, 2018
2. Nā haʻiʻōlelo, nā papa (wikiō)
1)
2)
3)
4)
5)
3. Nā kumu punaewele
1)
2)
4)
5)
6)