Sample ordering after RandomUnderSampler

[alexHeu]

Hi,

I have a question regarding the resulting sample ordering after using the RandomUnderSampler.

Currently, I run the sampling as follows: X_resampled, y_resampled = sampler.fit_resample(X_train, y_train)

However, when looking at the result, I saw that all positive samples are placed after the negative ones: plt.plot(y_resampled):

Is this expected behavior? Intuitively, it would make sense to me to keep the original ordering intact.

Would the following way be appropriate to achieve this? sample_indices = np.sort(sampler.sample_indices_) X_resampled = X_train[sample_indices] y_resampled = y_train[sample_indices]

best regards Alex

[hayesall]

Would the following way be appropriate to achieve this?

I think your sample_indices trick could keep the ordering for undersampling.

Is this expected behavior? Intuitively, it would make sense to me to keep the original ordering intact.

• You can maintain an ordering for random undersampling.
• This could be done for random oversampling (e.g. repeat the same index n times)
• But it's extremely unclear with techniques like SMOTE that generate synthetic points (e.g. you'd have to define an index for points that did exist originally).

Therefore it's unlikely to provide a fixed ordering in general.

Do you have a use-case that strongly requires a fixed ordering?

[glemaitre]

I am closing since this is a usage question