ficheprobapartiel.py

Created by raph-couvert

Created on May 27, 2025

1.43 KB


FICHE RECAPITULATIVE

1) STATISTIQUES PAR CLASSES

Soit N = effectif total, ni = effectif de la i-eme classe, ci = centre de classe, Fi = FCC en fraction (0 a 1)

MEDIANE Me
 Me = a + w*(0.5 - Fprev)/(Fnext - Fprev)
  a = borne inferieure de la classe mediane
  w = largeur de la classe
  Fprev = FCC avant la classe
  Fnext = FCC apres la classe

MOYENNE xbar
 xbar = sum(ci*ni)/N

VARIANCE et ECART-TYPE
 Var = sum(ci^2*ni)/N - xbar^2
 sigma = sqrt(Var)

2) BERNOULLI ET BINOMIAL

Xi ~ Bern(p)
 E(Xi) = p
 Var(Xi) = p*(1-p)

X = sum Xi  ~ Binomial(n,p)
 E(X) = n*p
 Var(X) = n*p*(1-p)

3) APPROXIMATION PAR TCL

Pour n grand, X ~ Bin(n,p) approximable par Normal(E,Var)
 E = n*p
 Var = n*p*(1-p)

Z = (X - E)/sqrt(Var)
 P(X <= x) ~ Phi((x + 0.5 - E)/sqrt(Var))   // si correction de continuite
 P(X >= x) ~ 1 - Phi((x - 0.5 - E)/sqrt(Var))

4) INTERVALLE DE CONFIANCE POUR PROPORTION

phat = X/n
IC a 1-alpha : phat +/- z_{1-alpha/2} * sqrt(phat*(1-phat)/n)
 z_{1-alpha/2}=1.96 pour 95%

5) LOI EXPONENTIELLE ET ESTIMATEURS

Xi i.i.d. ~ Exponential(lambda)

Likelihood
 L(lambda) = lambda^n * exp(-lambda*sum(xi))

Log-likelihood
 lnL = n*ln(lambda) - lambda*sum(xi)

EMV (MLE)
 d(lnL)/d(lambda) = n/lambda - sum(xi) = 0
 --> lambda_hat = n / sum(xi) = 1/xbar

Biais de lambda_hat
 E(lambda_hat) = (n-1)/n * lambda

Correcteur sans biais
 lambda_tilde = n/(n-1) * lambda_hat

Convergence
 Var(lambda_hat) -> 0  =>  lambda_hat -> lambda

During your visit to our site, NumWorks needs to install "cookies" or use other technologies to collect data about you in order to:

With the exception of Cookies essential to the operation of the site, NumWorks leaves you the choice: you can accept Cookies for audience measurement by clicking on the "Accept and continue" button, or refuse these Cookies by clicking on the "Continue without accepting" button or by continuing your browsing. You can update your choice at any time by clicking on the link "Manage my cookies" at the bottom of the page. For more information, please consult our cookies policy.