3
$\begingroup$

enter image description here

From this video,

https://www.youtube.com/watch?v=fr5wML3y1Xk

I deduced that here, $a==b$ and $c<0$.

Here are my questions,

(1) What does it mean by "equal loss values of errors"?

(2) What happens when a-priori probabilities of $C_1$ and $C_2$ are equal or unequal?

(3) Can you please supply me some link of study materials?

$\endgroup$
3
  • 1
    $\begingroup$ Can you describe or copy the description of the plot/image? What do the ellipses we are seeing represent? Also, my guess for "equal loss values of errors" would be an output that gives the same error-value regardless of the actual label. So if $o$ is our output and $E$ is our error-function/cost-function: $E(o, C_1)==E(o,C_2)$ $\endgroup$
    – dimpol
    Nov 17 '16 at 11:27
  • $\begingroup$ @dimpol, see the edit. $\endgroup$
    – user366312
    Nov 17 '16 at 11:36
  • $\begingroup$ If the horizontal and vertical axes are assumed to be drawn on the same scale, then it is not the case that $a=b$, because these describe the spreads of the data along those axes and the spreads clearly differ. $\endgroup$
    – whuber
    Nov 17 '16 at 20:10
3
+25
$\begingroup$

Assuming that both axes are on the same scale, you have $a<b$ and $c<0$. The contour plot for a bivariate normal distribution with the specified covariance matrix is an ellipse of the form:

$$\begin{align} \text{const} &= \begin{bmatrix} x_a - \mu_a & x_b - \mu_b \end{bmatrix} \begin{bmatrix} a & c \\ c & b \end{bmatrix}^{-1} \begin{bmatrix} x_a - \mu_a \\ x_b - \mu_b \end{bmatrix} \\[12pt] &= \frac{1}{ab-c^2} \begin{bmatrix} x_a - \mu_a & x_b - \mu_b \end{bmatrix} \begin{bmatrix} b & -c \\ -c & a \end{bmatrix} \begin{bmatrix} x_a - \mu_a \\ x_b - \mu_b \end{bmatrix} \\[12pt] &= \frac{b (x_a - \mu_a)^2 - 2c (x_a - \mu_a) (x_b - \mu_b) + a (x_b - \mu_b)^2}{ab-c^2}, \\[6pt] \end{align}$$

where $\mu_a$ and $\mu_b$ are the means of the two variables. You can tell that $c<0$ because the ellipse slopes in the negative direction (i.e., from top-left to bottom-right). You can tell that $a<b$ because the ellipse is narrower on the horizontal axis (for variable $x_a$) than on the vertical axis (for variable $x_b$).

The remaining questions pertaining to the probabilities for the classification problem is unclear to me, and more specification of the problem is needed.

$\endgroup$
1
  • $\begingroup$ Kindly, write something on the relationship between $P(C_1)$ and $P(C_2)$. What happens when they are equal or unequal? $\endgroup$
    – user366312
    Jul 22 at 17:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.