Условие
The softmax function turns a list of scores (logits) into probabilities. Language models use it to choose the next word:
softmax(x)_i = e^(x_i) / (e^(x_1) + e^(x_2) + … + e^(x_n))
Large scores make e^x overflow, so first subtract the largest score from every score. The result is mathematically the same.
Implement softmax(logits).
Input
logits: A list of scores.
Output
Return the list of probabilities, in the same order. They are all between 0 and 1 and add up to 1.
Ограничения
1 ≤ length ≤ 1,000
-1,000 ≤ each score ≤ 1,000
Примеры
Ввод:
logits = [1.0, 2.0, 3.0]
Ожидаемый вывод:
expected = [0.090031, 0.244728, 0.665241]
Ввод:
std::vector<double> logits = {1.0, 2.0, 3.0};
Ожидаемый вывод:
std::vector<double> expected = {0.090031, 0.244728, 0.665241};
Ввод:
double[] logits = new double[] {1.0, 2.0, 3.0};
Ожидаемый вывод:
double[] expected = new double[] {0.090031, 0.244728, 0.665241};
Пояснение
e¹ ≈ 2.718, e² ≈ 7.389 and e³ ≈ 20.086 add up to 30.193, so the probabilities are about [0.090031, 0.244728, 0.665241].
Ввод:
logits = [1000.0, 1000.0]
Ожидаемый вывод:
expected = [0.5, 0.5]
Ввод:
std::vector<double> logits = {1000.0, 1000.0};
Ожидаемый вывод:
std::vector<double> expected = {0.5, 0.5};
Ввод:
double[] logits = new double[] {1000.0, 1000.0};
Ожидаемый вывод:
double[] expected = new double[] {0.5, 0.5};
Пояснение
Computing e^1000 directly overflows. Subtracting the maximum gives [0, 0], and e⁰ / (e⁰ + e⁰) = 0.5 for both.
Как подступиться
Сначала попробуйте сами, затем открывайте подсказки по одной.
Подсказка 1
Subtract the largest logit before exponentiating so the largest transformed score becomes zero and overflow is avoided.
Подсказка 2
Compute each exponent, sum them once, then divide every exponent by that total to get probabilities.
Подсказка 3
Make sure the returned values stay in the same order as the input and still sum to one after normalization.
Ожидаемая сложность
O(n) time, O(n) extra space
Ссылки
Это задание адаптировано для CodeUtility. Следуйте условию и ограничениям выше.