---
dataset_info:
  features:
  - name: id
    dtype: int64
  - name: _priority
    dtype: float64
  splits:
  - name: train
    num_bytes: 141371024
    num_examples: 8835689
  download_size: 64076411
  dataset_size: 141371024
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# dataproc5/tmp-danbooru2025-row-priorities

The repo contains priority scores by id for tag blancing task, as part of the dataproc5 pipeline.

```python
import unibox as ub
from dataproc5.pipelines.danbooru.nodes import prepare_balancing_tags, select_balanced_images, prepare_row_priorities

dbr_df = ub.loads("hf://trojblue/danbooru2025-metadata").to_pandas()
tag_counts_all_df = ub.loads("hf://dataproc5/metrics-danbooru2025-alltime-tag-counts").to_pandas()

tag_df = prepare_balancing_tags(dbr_df, tag_counts_all_df, key_col="id")
dbr_df = dbr_df.merge(tag_df, on="id", how="left")

priority_df = prepare_row_priorities(dbr_df, column_name="_priority", key_col="id")
dbr_df = dbr_df.merge(priority_df, on="id", how="left")

# Then you can sort by the new priority
dbr_df.sort_values("_priority", ascending=False, inplace=True)
```