Alle Herausforderungen
Mittel 0 gelöst 0% Erfolgsquote

Train a Neuron with Gradient Descent

Repeat forward pass, gradients and weight updates to teach a sigmoid neuron from examples.

Das lernst du

  • Repeat forward pass, gradients and updates
  • Tune a learning rate
  • Watch a neuron learn from examples

Problem

Übersetzung anzeigen

Gradient descent improves the weights a little at a time by moving against the gradient:

w ← w - learning_rate × dL/dw

Train a sigmoid neuron with stochastic gradient descent. For each epoch, go through the samples in the given order, and for each sample:

  1. z = sample · weights + bias, y = 1 / (1 + e^(-z))
  2. delta = 2 × (y - target) × y × (1 - y)
  3. weights[i] ← weights[i] - learning_rate × delta × sample[i] (for every i), then bias ← bias - learning_rate × delta

Implement train_neuron(samples, targets, weights, bias, learning_rate, epochs).

Input

samples: m training examples, each with n values.

targets: The m correct outputs (between 0 and 1).

weights, bias: The starting weights (n values) and bias.

learning_rate, epochs: The step size and how many passes over the data.

Output

Return the trained weights followed by the trained bias (n + 1 numbers).

Einschränkungen

1 ≤ m ≤ 50, 1 ≤ n ≤ 10, 1 ≤ epochs ≤ 1,000

0 < learning_rate ≤ 5

-5 ≤ each value ≤ 5

Beispiele

Beispiel 1
Eingabe:
samples = [
    [1.0]
]
targets = [1.0]
weights = [0.0]
bias = 0.0
learning_rate = 1.0
epochs = 1

Erwartete Ausgabe:
expected = [0.25, 0.25]
Eingabe:
std::vector<std::vector<double>> samples = {
    {1.0}
};
std::vector<double> targets = {1.0};
std::vector<double> weights = {0.0};
double bias = 0.0;
double learning_rate = 1.0;
long long epochs = 1;

Erwartete Ausgabe:
std::vector<double> expected = {0.25, 0.25};
Eingabe:
double[][] samples = new double[][] {
    new double[] {1.0}
};
double[] targets = new double[] {1.0};
double[] weights = new double[] {0.0};
double bias = 0.0;
double learning_rate = 1.0;
long epochs = 1;

Erwartete Ausgabe:
double[] expected = new double[] {0.25, 0.25};

Erläuterung

y = σ(0) = 0.5, delta = 2 × (0.5 - 1) × 0.25 = -0.25. Weight: 0 - 1 × (-0.25) × 1 = 0.25. Bias: 0 - 1 × (-0.25) = 0.25. Result: [0.25, 0.25].

Beispiel 2
Eingabe:
samples = [
    [0.0, 0.0],
    [0.0, 1.0],
    [1.0, 0.0],
    [1.0, 1.0]
]
targets = [0.0, 0.0, 0.0, 1.0]
weights = [0.0, 0.0]
bias = 0.0
learning_rate = 2.0
epochs = 500

Erwartete Ausgabe:
expected = [6.346558, 6.339863, -9.592092]
Eingabe:
std::vector<std::vector<double>> samples = {
    {0.0, 0.0},
    {0.0, 1.0},
    {1.0, 0.0},
    {1.0, 1.0}
};
std::vector<double> targets = {0.0, 0.0, 0.0, 1.0};
std::vector<double> weights = {0.0, 0.0};
double bias = 0.0;
double learning_rate = 2.0;
long long epochs = 500;

Erwartete Ausgabe:
std::vector<double> expected = {6.346558, 6.339863, -9.592092};
Eingabe:
double[][] samples = new double[][] {
    new double[] {0.0, 0.0},
    new double[] {0.0, 1.0},
    new double[] {1.0, 0.0},
    new double[] {1.0, 1.0}
};
double[] targets = new double[] {0.0, 0.0, 0.0, 1.0};
double[] weights = new double[] {0.0, 0.0};
double bias = 0.0;
double learning_rate = 2.0;
long epochs = 500;

Erwartete Ausgabe:
double[] expected = new double[] {6.346558, 6.339863, -9.592092};

Erläuterung

This is the logical AND gate. After 500 epochs both weights are large and positive and the bias is strongly negative, so only the input [1, 1] gives an output above 0.5.

So gehst du vor

Versuch es zuerst allein und öffne dann einen Hinweis nach dem anderen.

Hinweis 1

Process the training samples in order for each epoch, updating weights immediately after each example.

Hinweis 2

For each sample, compute the sigmoid output, then use its gradient to move weights and bias downhill.

Hinweis 3

Because updates are online, later samples in the same epoch see earlier updates.

Erwartete Komplexität

O(epochs × m × n) time, O(1) extra space

Referenzen

Diese Aufgabe wurde für CodeUtility angepasst. Beachte die oben genannten Vorgaben und Einschränkungen.

Die beste bestandene Lösung jedes Coders. Schnellere, sparsamere und besser bewertete Lösungen stehen weiter oben.

Noch keine bestandene Lösung. Löse sie und hol dir Platz eins.

Von anderen lernen

Community-Lösungen

Noch keine veröffentlichten Lösungen. Löse die Aufgabe und teile als Erster eine Lösung.