pandaspro

pandas with friendlier dataframe methods and Excel export / formatting helpers.

pip install pandaspro

See CHANGELOG.md for what changed in each version, and df.cpdhelp() for the dataframe shortcuts.

Data assembly: cpdBaseFrame and cpdDatabaseC

The two halves of assembling a project's data:

• cpdBaseFrame — dated system extracts. One folder of files named by date (SOB_20261001.csv); decorate a class and load a version with X(version="latest"). The data is produced elsewhere and only read here.
• cpdDatabaseC — hand-maintained lists. One folder of JSON files named by list (modules/clients.json) plus a browser editor; decorate a store class and one class per list. The data is maintained by people in the published page.

cpdDatabaseC

from mylists import MyLists, Clients

MyLists.publish()          # creates <folder>/mylists.html and modules/*.json (data is never overwritten)
MyLists.open()             # opens the editor in Edge / Chrome; "Open folder" -> pick <folder>

Clients.frame()            # rows as a DataFrame
MyLists.list()             # every module: id, title, group, live, rows, revision, saved_at
MyLists.check()            # one line per module: filled? live? same as its old source?

CountryGroups.import_source()   # rows from the old Excel into the module (not live yet)
CountryGroups.compare()         # empty frame = module and Excel agree
CountryGroups.set_live()        # from now on read_live() returns the module

mylists is the example shipped in pandaspro/cpddatabasec/example/: copy mylists/ into your project, rename the store, keep or replace the three example lists, add your own. Declaring a store and a list:

from pandaspro import cpdDatabaseC, cpdField as field

@cpdDatabaseC(name="MyLists", package="myproject", editor="mylists.html")
class MyLists:
    @classmethod
    def get_path(cls):                       # or path="..." in the decorator
        return os.environ["MYLISTS_DIR"]     # per machine, never a hard-coded path in the repo

@MyLists.module("clients", title="Clients", group="Reference", key=["code"])
class Clients:
    code = field(required=True, help="Short code, unique")
    name = field(required=True)
    tier = field(type="select", options=["A", "B", "C"])

Add a list = add a decorated class. Add a field = add a line (publish() refreshes the definition, rows stay). Change the page = pass editor_source="your.html" (start from a copy of pandaspro/cpddatabasec/editor.html).

Word reports: WordReport

Take Word reports apart into sections, paragraphs and tables, check that nothing was lost, put several together and write one .docx with a consistent layout. Standard library only.

import pandaspro as cpd

a = cpd.WordReport.read("part_a.docx")
b = cpd.WordReport.read("part_b.docx")

a.preview()                 # numbered outline: S1 sections, P3 paragraphs, T2 tables
a.preview_html()            # rendered page in the browser, with the check on top
print(a.check())            # characters, digits, every number, footnotes, tables vs the source

table = a.tables[0]
table.title, table.header_rows, table.notes, table.footnotes
table.frame                 # DataFrame; table.to_frame(numeric=True) parses the numbers

merged = cpd.WordReport.merge(a, b, titles=["Part A", "Part B"])
merged.to_docx("out/combined.docx")   # written, read back and compared; raises on any difference

What is kept: the text, bold / italic / underline / strike, colour, highlight, super and subscript, links, footnotes and endnotes, bullets and numbering, merged cells, header rows, cell shading, pictures and text boxes. Fonts and font sizes are dropped on purpose: the output uses one layout. Charts and diagrams cannot be carried; a red placeholder marks their position and report.warnings lists them. The table of contents, page headers / footers and comments are not read.

Table titles ("Table 1 ...", the Caption style) and notes ("Note:", "Source:", "a/ ...") are found by rule. When a rule guesses wrong, pass title_pattern= / note_pattern= to read(), or move the paragraphs yourself: section.blocks, table.title and table.notes are ordinary lists.

report.to_dict() / to_json() / WordReport.from_json() expose the whole structure.

Some Word files lay a whole page out as one table: a row of running text, empty spacer rows, a "Table 1 ..." row, the real header and data rows, another title row. read() takes such layout tables apart into paragraphs and separate tables and drops the empty rows and columns (unpack_layout=False keeps them as they are; report.unpacked says how many were found).

Merging

merged = cpd.WordReport.merge(a, b, tables_last=True, tables_heading="Tables", number_tables=True)

tables_last puts the text of all reports first (headings and paragraphs, in order) and all their tables after it. number_tables gives the table titles "Table 1.", "Table 2." ... in that order: a number a title already has ("Table A1:") is replaced, a table without a title gets "Table N"; nothing else in the titles changes. report.number_tables(prefix="Table", sep=".") does the same on any report.

Reviewing before writing

merged.review_html()                     # opens a page: keep / drop, rename, title and subtitle
merged = merged.apply_review(r'''{...}''')   # the line the page's "Copy result" button gives

Every section, paragraph and table on the page has a checkbox; headings and table titles are editable; a title and subtitle can be typed at the top. Cells and numbers cannot be changed there. apply_review() takes the JSON as a dict, a string or the path of the downloaded review.json and returns a new report; check() still passes, with the dropped text counted.

Writing into a template

to_docx(template=...) writes into your own blank .docx / .dotx: its page header and footer, page size and margins, theme and styles are kept, and the report goes under whatever its body holds (a title line, empty lines that leave room for a banner). keep_template_body=False starts from an empty body. fill={"XXX": "Region A", "MM DD, YYYY": "September 30, 2026"} replaces placeholders in the template body, page headers and footers, also when Word split them over several runs, and removes their highlight. A placeholder the template does not have gives a warning (fill_missing="error" raises, "ignore" stays quiet). WordReport.inspect_template(path) lists the template's style names, its highlighted placeholders and its body text. Headings, captions and tables take the template's styles of the same name ("heading 1", "Table Grid" ...). styles= points each part at another style or changes it:

merged.to_docx(
    "out/combined.docx",
    template="blank_template.docx",
    font="Arial",                      # body font; left out, the template's stays
    styles={
        "heading1": "Report Heading",  # a style name in the template
        "heading2": {"font": "Georgia", "size": 13, "color": "1F4E79", "space_before": 18},
        "body": {"east_asia_font": "SimSun", "line_spacing": 1.15},
        "table": {"style": "Grid Table 4", "header_fill": "DDEBF7", "borders": "horizontal"},
        "table_header": {"bold": True},
    },
)

Roles: body, heading1 … heading9, title, subtitle, list, quote, caption, table_title, table_text, table_header, table_note, footnote, endnote, table. Paragraph options: font, east_asia_font, size, bold, italic, underline, color, all_caps, small_caps, letter_spacing (points, negative is narrower), align, space_before / space_after (points), line_spacing, indent / first_line (inches), keep_with_next, page_break_before, and style to start from a named style. Table options: style, borders (grid, horizontal, outside, none), border_color, border_size, header_fill (a colour, or none for no shading), header_color (text in header rows), header_height / row_height (inches), align (the table on the page), header_align / text_align / first_column_align (the cells: header rows, the rest, the first column in every row), even_columns (rest: every column but the first gets the same width; all), first_column, banded_rows, keep_together (default on). An option that sets bold, italic or underline on a role also overrides what the source runs carried. A name the template does not have raises an error that lists the names it does have. page=, landscape= and margin= override the template's page setup only when given.

A table stays on one page (it moves to the next page whole when it does not fit; no page break is inserted). blank_lines=True sets paragraphs apart with an empty line (none under a heading, inside a list or after a table); margin_top="auto" starts the text under the page header's banner on every page, with header_gap inches of air under it (0.1 by default); new_page=r"^annex" makes the headings that match begin a new page; number_tables=True numbers the table titles. preset= takes these five and styles together as one dict, with styles= laid over it; the role headings stands for heading1 … heading9. title= / subtitle= add those two lines above the report:

merged.to_docx("out/brief.docx", template="blank_template.docx", preset={...},
               title="Regional Brief", subtitle="As of September 30, 2026",
               fill={"XXX": "Region A", "MM DD, YYYY": "September 30, 2026"})

Profiles: a project's rules in one file

Nothing project-specific is built in. What a project knows about its own reports — which lines are headings, what to leave out, the merge order, the fonts and table look — goes in one YAML (or JSON) file, kept next to the project, and is passed as profile=:

profile = cpd.WordReport.profile("project.yaml")
a = cpd.WordReport.read("Note on Region A.docx", profile=profile)   # the matching document's rules
b = cpd.WordReport.read("Outcomes note.docx", profile=profile)
merged = cpd.WordReport.merge(a, b, profile=profile)                # order, headings, tables last
merged.review_html()                                                # optional, as above
merged.to_docx("out/brief.docx", profile=profile, fill={"XXX": "Region A"})
report = profile.combine(["Note on Region A.docx", "Outcomes note.docx"])   # read + merge in one go

name: Regional note
documents:                        # one entry per kind of source file
  first:
    name: First Team
    match: '^Note on\b'           # tried on the file name, then on the first lines of text
    heading: ~                    # a text here puts the whole document under that heading
    drop: ['^Summary of\b']       # paragraphs to leave out (counted as read by check())
    headings:
      - {italic_line: true, max_chars: 100, level: 1}   # a line in italics throughout
      - {match: '^ANNEX\s*:?\s*$', text: Annex, level: 1}   # rewritten as "Annex"
    table_titles: ['^(Headcount|Offices)\s*\(FY\d{2}']  # the paragraph above a table
  second:
    name: Second Team
    match: '^Outcomes\b'
    drop: ['^Confidential\b']
    headings:
      - {match: '^Outcomes and Promotion Note\b', text: Promotion, level: 1}
merge:
  order: [first, second]
  tables_last: true
  tables_heading: Annex           # a source's own emptied "Annex" section is reused
  number_tables: true
write:                            # everything to_docx takes: styles, blank_lines, margin_top,
  blank_lines: true               # new_page, number_tables, template (relative to this file),
  margin_top: auto                # font, size, page, landscape, margin, table_size, note_size,
  new_page: '^\s*(annex|appendix)\b'   # placeholders, keep_template_body, fill_missing
  number_tables: true
  styles:
    body:     {space_before: 0, space_after: 0, line_spacing: 1.0}
    title:    {font: Open Sans, size: 14, bold: true, align: center, space_after: 8}
    headings: {font: Open Sans, size: 11, align: left}
    heading1: {italic: true}
    table:    {header_fill: '156082', header_color: 'FFFFFF', header_height: 0.5}

read() picks the document whose match fits the file name or one of its first three lines (document="first" names it outright); report.document_name says which one was used. The rules turn paragraphs into headings (the heading style decides their look), leave out the ones in drop, and add the document's own table title and note patterns to the defaults. merge() puts the reports in order, under the documents' heading when one is given, and fills in tables_last, tables_heading and number_tables; what is passed to merge() wins. to_docx() takes the write section as its defaults, again under what is passed. Patterns are regular expressions searched in the text; match, drop and headings are case-sensitive, table titles and notes are not. A key the profile does not know raises an error that lists the known ones.

Per table, by its title: tables: [{title: '^Staff\b', corner: Staff, shade: none}] sets the text of the top-left cell and which body rows are shaded (shade: none, or patterns matched on the first cell of a row: those rows are shaded whole, the others not; shade_fill picks the colour, else the one the table used).

Several files from one merged report go in outputs:; profile.write_outputs(report, folder, stem, template=..., title=...) writes them all (report.tables_only() is what content: tables uses):

outputs:
  internal:
    file: '{stem} - Internal.docx'
    content: all
    stamp: {find: Strictly Confidential, text: Internal Use Only, fill: FFC000}
  tables:
    file: '{stem} - Tables.docx'
    content: tables                 # nothing but the tables, with their titles and notes
    subtitle: Tables

stamp (also to_docx(stamp=...)) changes a label of the template: the text, whatever its case and however Word split it, and the background of the text box or shape that holds it (a plain label gets the colour behind its runs). Either text or fill may be left out.

The page

cpd.WordReport.app()                 # in a notebook opened in the project folder

opens a local page that does the whole job without code: Quick read finds the profile, the template and the source files in the folder (files are sorted to their documents by name), or drop a profile and the files on the page; Read shows each document's outline and check and the merged one; Review is the review page, inline; Generate writes the outputs and opens them. The page talks to a small server on 127.0.0.1 inside the Python process, so the kernel has to stay alive; nothing leaves the machine, dropped files are copied to a temporary folder. app(profile=..., folder=..., block=True) for a script; the returned object has .url and .stop().

Versions and releases

All work happens on master. A version is a tag on master (v1.3.1); there are no version branches. The version number lives in one place, pyproject.toml.

• While you work: add a line under Unreleased in CHANGELOG.md for anything a user can notice.
• To release: python release.py 1.3.2 (add --dry-run to rehearse). It checks the tree, runs the tests, sets the version, dates the changelog section, commits, tags and pushes. The pushed tag makes GitHub Actions build the package and publish it to PyPI.
• Every push to master runs the tests on GitHub.
• A machine that produces reports should sit on a tag, not on every push: git fetch --tags then git checkout v1.3.2.
• Version numbers: last digit for fixes and small additions, middle digit for new features, first digit when existing calls stop working.