Skip to content

Update README.md - #9

Merged
monica-cheng1 merged 1 commit into
mainfrom
monica-cheng1-patch-1
Aug 25, 2026
Merged

monica-cheng1 merged 1 commit into
mainfrom
monica-cheng1-patch-1

Conversation

@monica-cheng1

@monica-cheng1 monica-cheng1 commented Aug 25, 2026 •

Copy link
Copy Markdown
Contributor

Greptile Summary

The PR replaces the two-line README with detailed user-facing documentation for the block’s purpose, clustering pipeline, inputs and outputs, use cases, comparisons, FAQ, citations, and ecosystem links.

  • Embedding clustering — grouping sequences using distances between protein-language-model vectors rather than direct residue identity. The PR makes this the README’s central explanation and documents its use cases.
  • Centered PCA — dimensionality reduction after centering data and selecting principal components. The PR documents a 95% variance target, though the implementation may retain less when component caps bind.
  • L2 normalization — scaling each reduced embedding to unit Euclidean length. The PR newly describes it as the preprocessing step after PCA.
  • HDBSCAN — density-based clustering that discovers cluster counts and permits noise points. The PR explains its role and parameters, but incorrectly attributes the implementation to scikit-learn rather than the standalone hdbscan package.
  • Minimum cluster size — the smallest group HDBSCAN can recognize as a cluster. The PR introduces it as the main user-facing granularity control.
  • Noise rescue — a second clustering attempt over initially unassigned points. The PR documents that it is enabled by default and that remaining points become singleton clusters.
  • Large-cluster refinement — recursively splitting unusually large clusters when meaningful subclusters are found. The PR adds an explanation of this refinement.
  • Medoid — a real cluster member selected as its representative. The PR introduces and defines this term, although the definition does not precisely describe the exact weighted-medoid calculation.

Confidence Score: 4/5

The documentation-only PR appears safe to merge after addressing several non-blocking technical-accuracy issues.

The README broadly matches the current block, but it names the wrong HDBSCAN implementation, overstates the PCA variance guarantee, and simplifies the medoid definition in a way that can mislead readers reproducing the pipeline.

Files Needing Attention: README.md

Important Files Changed

Filename Overview
README.md Adds comprehensive product and algorithm documentation; most capability claims match the implementation, but the HDBSCAN backend, retained-variance guarantee, and medoid definition need correction.

Fix all with Greploop Fix All in Claude Code

Prompt To Fix All With AI
### Issue 1
README.md:29
**Wrong HDBSCAN implementation attribution**

The specification and citation attribute clustering to scikit-learn, but the block imports and pins the standalone `hdbscan` package. Readers following this documentation can reproduce or cite a different implementation from the one the block actually runs.

### Issue 2
README.md:30
**PCA variance target overstated**

This presents 95% retained variance as a fixed preprocessing result, but input-size and configured component caps can bind before that threshold is reached. Describing 95% as a target would prevent users from overestimating the dimensionality-reduction fidelity of large runs.

### Issue 3
README.md:79-81
**Medoid definition mismatches calculation**

The exact medoid path minimizes probability-weighted distances to other cluster members rather than selecting the member nearest the cluster center. These criteria can choose different representatives, so the definition should describe the implemented weighted-medoid calculation.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Reviews (1): Last reviewed commit: "Update README.md" | Re-trigger Greptile

Greptile also left 3 inline comments on this PR.

Context used:

@monica-cheng1
monica-cheng1 merged commit ff0b541 into main Aug 25, 2026
2 checks passed
Comment thread README.md
| | |
|---|---|
| Block title in app | Embedding Clustering |
| Algorithm | HDBSCAN (scikit-learn), density-based — cluster count discovered, not specified |

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Wrong HDBSCAN implementation attribution

The specification and citation attribute clustering to scikit-learn, but the block imports and pins the standalone hdbscan package. Readers following this documentation can reproduce or cite a different implementation from the one the block actually runs.

Knowledge Base Used: Embedding clustering computation

Prompt To Fix With AI
This is a comment left during a code review.
Path: README.md
Line: 29

Comment:
**Wrong HDBSCAN implementation attribution**

The specification and citation attribute clustering to scikit-learn, but the block imports and pins the standalone `hdbscan` package. Readers following this documentation can reproduce or cite a different implementation from the one the block actually runs.

**Knowledge Base Used:** [Embedding clustering computation](https://app.greptile.com/milaboratories/-/custom-context/knowledge-base/platforma-open/embedding-clustering/-/docs/embedding-clustering-computation.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

Fix in Claude Code

Comment thread README.md
|---|---|
| Block title in app | Embedding Clustering |
| Algorithm | HDBSCAN (scikit-learn), density-based — cluster count discovered, not specified |
| Preprocessing | Centered PCA to 95% variance, then L2 normalization |

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 PCA variance target overstated

This presents 95% retained variance as a fixed preprocessing result, but input-size and configured component caps can bind before that threshold is reached. Describing 95% as a target would prevent users from overestimating the dimensionality-reduction fidelity of large runs.

Knowledge Base Used: Embedding clustering computation

Prompt To Fix With AI
This is a comment left during a code review.
Path: README.md
Line: 30

Comment:
**PCA variance target overstated**

This presents 95% retained variance as a fixed preprocessing result, but input-size and configured component caps can bind before that threshold is reached. Describing 95% as a target would prevent users from overestimating the dimensionality-reduction fidelity of large runs.

**Knowledge Base Used:** [Embedding clustering computation](https://app.greptile.com/milaboratories/-/custom-context/knowledge-base/platforma-open/embedding-clustering/-/docs/embedding-clustering-computation.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

Fix in Claude Code

Comment thread README.md
Comment on lines +79 to +81
### What is the medoid for?

It is the member nearest the cluster's center — the natural representative when you want one real sequence to stand for a family, for ordering, expression, or reporting.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Medoid definition mismatches calculation

The exact medoid path minimizes probability-weighted distances to other cluster members rather than selecting the member nearest the cluster center. These criteria can choose different representatives, so the definition should describe the implemented weighted-medoid calculation.

Knowledge Base Used: Embedding clustering computation

Prompt To Fix With AI
This is a comment left during a code review.
Path: README.md
Line: 79-81

Comment:
**Medoid definition mismatches calculation**

The exact medoid path minimizes probability-weighted distances to other cluster members rather than selecting the member nearest the cluster center. These criteria can choose different representatives, so the definition should describe the implemented weighted-medoid calculation.

**Knowledge Base Used:** [Embedding clustering computation](https://app.greptile.com/milaboratories/-/custom-context/knowledge-base/platforma-open/embedding-clustering/-/docs/embedding-clustering-computation.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

Fix in Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant