Lära Density-Weighted Sampling

Density-weighted sampling is a strategy in active learning that helps you select the most valuable data points for labeling. Unlike pure uncertainty sampling, which focuses only on how uncertain the model is about each sample, density-weighted sampling also considers how representative a sample is within the data distribution. The intuition is simple: you want to prioritize not just uncertain points, but also those that are typical of the dataset, avoiding rare outliers that may not help the model generalize. By combining informativeness (such as uncertainty) with sample density, you can focus your labeling effort on data points that both challenge the model and represent common patterns in your data.


              12345678910111213141516171819202122232425262728293031
            
import numpy as np
from sklearn.neighbors import NearestNeighbors
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier

# Generate a synthetic dataset
X, y = make_classification(n_samples=200, n_features=2, n_informative=2, 
                           n_redundant=0, n_clusters_per_class=1, random_state=42)

# Train a simple classifier
clf = RandomForestClassifier(random_state=42)
clf.fit(X, y)

# Compute uncertainty: use predicted class probabilities
probs = clf.predict_proba(X)
uncertainty = 1 - np.max(probs, axis=1)  # Least confident score

# Estimate sample density using k-nearest neighbors
k = 5
nbrs = NearestNeighbors(n_neighbors=k+1)  # +1 because the point itself is included
nbrs.fit(X)
distances, _ = nbrs.kneighbors(X)
density = 1 / (np.mean(distances[:, 1:], axis=1) + 1e-10)  # Avoid division by zero

# Combine uncertainty and density (simple product)
density_weighted_score = uncertainty * density

# Select the top 10 samples by density-weighted score
top_indices = np.argsort(-density_weighted_score)[:10]

print("Indices of top 10 density-weighted samples:", top_indices)

Var allt tydligt?

Tack för dina kommentarer!

Avsnitt 2. Kapitel 4

Fråga AI

Fråga vad du vill eller prova någon av de föreslagna frågorna för att starta vårt samtal

Suggested prompts:

Can you explain how the density is calculated in this example?

What does the uncertainty score represent in this context?

How would you adjust this approach for a different dataset or model?

Svep för att visa menyn


              12345678910111213141516171819202122232425262728293031
            
import numpy as np
from sklearn.neighbors import NearestNeighbors
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier

# Generate a synthetic dataset
X, y = make_classification(n_samples=200, n_features=2, n_informative=2, 
                           n_redundant=0, n_clusters_per_class=1, random_state=42)

# Train a simple classifier
clf = RandomForestClassifier(random_state=42)
clf.fit(X, y)

# Compute uncertainty: use predicted class probabilities
probs = clf.predict_proba(X)
uncertainty = 1 - np.max(probs, axis=1)  # Least confident score

# Estimate sample density using k-nearest neighbors
k = 5
nbrs = NearestNeighbors(n_neighbors=k+1)  # +1 because the point itself is included
nbrs.fit(X)
distances, _ = nbrs.kneighbors(X)
density = 1 / (np.mean(distances[:, 1:], axis=1) + 1e-10)  # Avoid division by zero

# Combine uncertainty and density (simple product)
density_weighted_score = uncertainty * density

# Select the top 10 samples by density-weighted score
top_indices = np.argsort(-density_weighted_score)[:10]

print("Indices of top 10 density-weighted samples:", top_indices)

Var allt tydligt?

Tack för dina kommentarer!

Avsnitt 2. Kapitel 4