Problema
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.
Restrições
1 ≤ length ≤ 1,000
-1,000 ≤ each score ≤ 1,000
Exemplos
Entrada:
logits = [1.0, 2.0, 3.0]
Saída esperada:
expected = [0.090031, 0.244728, 0.665241]
Entrada:
std::vector<double> logits = {1.0, 2.0, 3.0};
Saída esperada:
std::vector<double> expected = {0.090031, 0.244728, 0.665241};
Entrada:
double[] logits = new double[] {1.0, 2.0, 3.0};
Saída esperada:
double[] expected = new double[] {0.090031, 0.244728, 0.665241};
Explicação
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].
Entrada:
logits = [1000.0, 1000.0]
Saída esperada:
expected = [0.5, 0.5]
Entrada:
std::vector<double> logits = {1000.0, 1000.0};
Saída esperada:
std::vector<double> expected = {0.5, 0.5};
Entrada:
double[] logits = new double[] {1000.0, 1000.0};
Saída esperada:
double[] expected = new double[] {0.5, 0.5};
Explicação
Computing e^1000 directly overflows. Subtracting the maximum gives [0, 0], and e⁰ / (e⁰ + e⁰) = 0.5 for both.
Como abordar
Tente sozinho primeiro e depois abra uma dica de cada vez.
Dica 1
Subtract the largest logit before exponentiating so the largest transformed score becomes zero and overflow is avoided.
Dica 2
Compute each exponent, sum them once, then divide every exponent by that total to get probabilities.
Dica 3
Make sure the returned values stay in the same order as the input and still sum to one after normalization.
Complexidade esperada
O(n) time, O(n) extra space
Referências
Este desafio foi adaptado para o CodeUtility. Siga o enunciado e as restrições acima.