Matlab 2015Ra绘制大型数组散点图内存占用过高求助

Hey there, I get it—trying to plot 21 million+ points with scatter() is a surefire way to eat up all your memory, especially in older MATLAB versions that don't have binscatter. Let's break down practical, memory-friendly alternatives you can use right now:

1. Downsample Your Data (Quickest Fix) #

With that many points, your eye can't distinguish individual markers anyway. Randomly sample a smaller subset that still captures the overall distribution:

a = randn(21026304,1);
% Adjust the sample ratio based on how much detail you need (0.01 = 1% of points)
sample_ratio = 0.01;
sample_indices = randperm(length(a), round(length(a)*sample_ratio));
scatter(a(sample_indices), a(sample_indices), 'MarkerSize', 1);

This cuts your data size drastically while preserving the core pattern of your dataset.

2. Use 2D Density Plots Instead of Scatter #

Instead of rendering every single point, plot the density of points in bins—this is way more memory-efficient and often clearer for large datasets. Try histogram2 (available in R2015a):

a = randn(21026304,1);
% Adjust the number of bins to control detail vs. performance
histogram2(a, a, 200, 'DisplayStyle', 'tile', 'ShowEmptyBins', 'off');
colorbar; % Add color scale to show point density
xlabel('a'); ylabel('a');

If you prefer a heatmap-style view, use histcounts2 with imagesc:

[N, x_edges, y_edges] = histcounts2(a, a, 200);
imagesc(x_edges, y_edges, N');
axis xy; % Flip axis to match scatter plot orientation
colorbar;
xlabel('a'); ylabel('a');

3. Swap scatter() for plot() with Markers #

The scatter() function is memory-heavy because it lets you customize each marker individually. For uniform markers, plot() uses far less memory:

a = randn(21026304,1);
plot(a, a, '.', 'MarkerSize', 1); % Use tiny dots to avoid clutter

This renders all points in one go with minimal overhead, and it's almost as fast as plotting a regular line.

4. Plot in Chunks #

If you absolutely need to visualize every point, split your dataset into smaller chunks and plot them sequentially. This prevents MATLAB from loading all data into the graphics memory at once:

a = randn(21026304,1);
chunk_size = 1e6; % Process 1 million points at a time
figure; hold on;
for idx = 1:chunk_size:length(a)
    end_idx = min(idx + chunk_size - 1, length(a));
    plot(a(idx:end_idx), a(idx:end_idx), '.', 'MarkerSize', 1);
end
hold off;

You might notice a slight slowdown from looping, but the memory savings are massive.

Quick Recommendation #

If your goal is to understand the data's distribution, go with the density plot method—it's the most efficient and informative. If you need to show individual points (e.g., for outlier detection), downsampling or chunked plotting will work best.

内容的提问来源于stack exchange，提问作者Kanjoo