Data Format Bridge

Data format bridge (SPSS / R / Prism / Excel / Stata)

Use this skill whenever a result table has to leave Python: the user says "导出成 SPSS 能打开的", "给我 Prism 能画图的表", "整理成 Excel", "我想在 R 里接着分析", "Stata 文件", or an analysis skill has produced CSVs that a clinician will take to their own statistics software. Also use inspect to look inside files the user brings (.sav, .dta, .rds, .xlsx, .h5ad) before doing anything with them.

Target mapping — what the user says → what you write:

Commands

bashCopy code
python "<skill dir>/convert.py" inspect --input results/de_table.csv [--sheet obs]
python "<skill dir>/convert.py" convert --input results/de_table.csv --to sav
python "<skill dir>/convert.py" convert --input results/scores.csv --to pzfx --group-col group --value-col score
python "<skill dir>/convert.py" convert --input results/scores.csv --to r [--columns patient,group,score]
python "<skill dir>/convert.py" convert --input results/book.xlsx --to xlsx --sheet Sheet2 --out exports/clean.xlsx
python "<skill dir>/convert.py" h5ad-tables --input data/adata.h5ad --out exports/adata_tables --genes CD8A,GZMB,PDCD1

• inspect prints shape, columns, dtypes, first 5 rows, missing counts (and SPSS/Stata variable labels when present). For .h5ad it reads only obs/var (no expression matrix); use --sheet obs|var.
• convert writes into ./exports/ under the current workspace unless --out is given; existing files are never overwritten (a _1, _2 suffix is added and reported in notes). The input file is never modified, and --out equal to the input is refused.
• --group-col/--value-col pivots a long table (one row per measurement) into one column per group — this is what Prism's Column tables and t-test/ANOVA dialogs expect. Without it, pzfx uses every numeric column as a group and reports what it dropped.
• h5ad-tables writes obs.csv, var.csv, and with --genes an obs_with_genes.csv (per-cell metadata + expression of those genes, taken as stored in X or --layer). Feed that file back into convert --to sav|pzfx|xlsx for per-cell / per-group statistics outside Python.

Read the JSON output and relay it

Every run reports files, columns_out, and for SPSS/Stata renamed_columns (SPSS names must be ≤64 bytes, letters/digits/underscore, no leading digit, no reserved words like ALL; Stata ≤32 chars) plus type_changes (bool → 0/1, categorical → string, timezone dropped). Tell the user which columns were renamed — the original names are preserved as variable labels so they still appear in SPSS "Variable View" / Stata describe.

After converting, always tell the user the absolute file path and how to open it:

• SPSS: 文件(File) → 打开(Open) → 数据(Data)，选择 .sav；原列名在“变量标签”中。
• Prism: File → Open 打开 .pzfx；如果 Prism 报错或版本较旧，新建 Column 表后 File → Import 导入 *_prism_wide.csv（CSV/XLSX 导入是最稳妥的路线，.pzfx 是按 Prism 8 的 XML 结构手写的最小文件）。
• R: 打开 *_load.R，逐行运行；或 df <- readRDS("<path>")。
• Excel: 直接双击；Stata: File → Open 或 use "<path>", clear。

Environment

Reading needs only pandas. Writers are imported lazily: pyreadstat (sav/dta), pyreadr (rds/r), openpyxl (xlsx), anndata (h5ad). If a run errors with "package ... is not installed", run the uv pip install <pkg> command from the message and retry; the BioDSH env pack “统计与临床” installs pyreadstat and openpyxl (add pyreadr with uv pip install pyreadr).

Rules

• Never overwrite or edit the input; never write outside exports/ unless the user asked for a specific --out.
• Do not "fix" the data while converting (no dropping rows, no recoding) — if the table needs cleaning, do it as a separate, explained step and keep the original.
• Missing values stay missing (empty cell in Prism/Excel, system-missing in SPSS/Stata, NA in R); say so if the user asks about blanks.