Linear Regression Is Too Easy 1

Yulim is confident about linear regression. During a MatKor club lecture on the topic, she was not paying attention. Dongwoo, who was the instructor at the time, did not like this, so he gave her a harder problem. In ordinary linear regression, given data $(x_1, y_1), \dots, (x_n, y_n)$, you find real numbers $a_2$ and $b_2$ that minimize $\sum_{i=1}^n (a_2 x_i + b_2 - y_i)^2$, the sum of squared residuals. Dongwoo extended this to the sum of residuals raised to the power $k$, $\sum_{i=1}^n (a_k x_i + b_k - y_i)^k$, and asked for real numbers $a_k$ and $b_k$ that make this sum as close to $0$ as possible. Yulim found this too hard and asked for something easier. Dongwoo thought about it and added two conditions. Only the case $k = 1$ is asked, and the $y$-intercept is fixed so that only the slope must be found.

Help Yulim. For a given $b$, find the real number $a_1$ that makes $\sum_{i=1}^n (a_1 x_i + b - y_i)^1$ as close to $0$ as possible.

The first line contains an integer $n$ ($1 \le n \le 10^5$), the number of data points, and an integer $b$ ($-10^9 \le b \le 10^9$), the $y$-intercept. Each of the next $n$ lines contains one point, given by integers $x_i$ and $y_i$ ($-10^9 \le x_i, y_i \le 10^9$). The same point may appear more than once.

On the first line, print the value of $a_1$ that makes $\sum_{i=1}^n (a_1 x_i + b - y_i)^1$ as close to $0$ as possible.

If the answer is an integer, print it as is. If the answer is a rational number that is not an integer, reduce it to an irreducible fraction $p/q$ ($1 \le |p|$, $2 \le q$, $\gcd(|p|, q) = 1$) and print it in the form p/q. It can be proven that a rational $a_1$ achieving the closest value to $0$ always exists.

If more than one $a_1$ is possible, print EZPZ.