Run any Skill in Manus with one click

pyfixest-cupy64-absorbed-regressors

Stars1

Forks0

UpdatedJuly 22, 2026 at 09:34

pyfixest demeaner_backend="cupy64" (including its CPU fallback when cupy is absent) is NOT numerically identical to the default numba backend and does NOT drop fully-absorbed/collinear regressors the same way. Use when: (1) adding demeaner_backend="cupy64" to existing pf.feols/fepois calls changes the printed coefficient table, (2) a regression report suddenly gains rows with absurd estimates (e.g. coef 435.8, SE 7106) for controls absorbed by the fixed effects, (3) diffing outputs before/after a backend change, or (4) anything parses a pyfixest text report by line position.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

kennethkhoocy

kennethkhoocy/applied-micro-skills

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

File Explorer

2 files

SKILL.md

readonly

name	pyfixest-cupy64-absorbed-regressors
description	pyfixest demeaner_backend="cupy64" (including its CPU fallback when cupy is absent) is NOT numerically identical to the default numba backend and does NOT drop fully-absorbed/collinear regressors the same way. Use when: (1) adding demeaner_backend="cupy64" to existing pf.feols/fepois calls changes the printed coefficient table, (2) a regression report suddenly gains rows with absurd estimates (e.g. coef 435.8, SE 7106) for controls absorbed by the fixed effects, (3) diffing outputs before/after a backend change, or (4) anything parses a pyfixest text report by line position.
author	Claude Code
version	1.0.0
date	"2026-07-18T00:00:00.000Z"

pyfixest cupy64 backend: absorbed regressors survive, reports change shape

Problem

Adding demeaner_backend="cupy64" to an existing pf.feols() call is treated as a pure performance switch, but it changes the output: regressors with zero within-FE identifying variation (fully absorbed by the fixed effects) that the default numba backend silently drops are RETAINED by the cupy64 path (also in its scipy/CPU fallback when cupy is not installed). They appear in the coefficient table as non-identified garbage (huge coefficient, huge SE).

Context / Trigger Conditions

Observed 2026-07-18 (Specialist Directors US, H5 re-baseline audit): adding the kwarg to 4 feols sites left the headline triple stable to 4 decimals (B1 diff 1.45e-6, SE diff 9.9e-6) but the text report grew from 366 to 390 lines — 24 new rows for absorbed controls (e.g. event_x_lrisk = 435.8059, SE 7106.3365) under firm-year + director FE, where firm-year-level variables have no identifying variation.

Solution

Treat a backend change as a POTENTIALLY OUTPUT-CHANGING edit: diff the report and expect schema changes, not byte equality. Verify the coefficients of interest at ~4-decimal precision instead.
Never interpret retained absorbed-regressor rows as estimates; check within-FE variation before reading nuisance coefficients.
Never parse pyfixest text reports by line position; anchor on the variable name of the coefficient you need.
When byte-stable reports matter (regression-tested pipelines), pin the backend consistently everywhere rather than mixing backends across runs.

Verification

Re-run one model with and without the kwarg; compare: headline coef equal to ~1e-6, coefficient-row sets DIFFERENT (absorbed regressors present only under cupy64). That asymmetry confirms this behavior rather than a data change.

Notes

Applies even with no GPU: the fail-open CPU fallback shows the same retention behavior, so "cupy isn't installed" does not make the kwarg inert.
The headline inference (identified coefficients, clustered SEs) agrees to reporting precision; this is a report-shape/nuisance-row issue, not a correctness issue for identified estimates.

More from this repository

same repository

adjudication-sheets

kennethkhoocy/applied-micro-skills

Build human adjudication / hand-labeling sheets from LLM-pipeline data without evidence truncation. Use when: (1) preparing a CSV/Excel sheet for a human to rule on cases an LLM classifier or rater panel judged, (2) a labeler reports "there is no information to label from" or cells look empty in Excel, (3) excerpt columns cluster at one exact length (e.g. all 1,500 chars — a hard truncation cap). Covers: full rating-basis recovery, Excel 32,767-char cell cap, multi-line CSV mangling, ruling dropdowns, companion text files.

2026-07-221

adversarial-empirical-review

kennethkhoocy/applied-micro-skills

N-round adversarial review pipeline for empirical research output — the chain from data to LaTeX tables to a manuscript that cites them. A Claude drafter proposes minimal diffs, a deterministic mechanical battery gates every diff from a clean state with a regression gate, a Codex reviewer files check-backed critiques, and a blind judge panel decides residual disputes. Manual-invoke ONLY: trigger when the user explicitly runs /adversarial-empirical-review or names 'adversarial-empirical-review' / 'adversarial empirical review'. Do NOT auto-trigger on generic 'review my results', 'check my tables', or manuscript-editing requests. For prose-style refinement use style-emulation instead; this skill AUDITS WHETHER THE TABLES ARE CORRECT — that each number in the tables is what the analysis code computes, reproduces from the data, and is internally consistent. It is an empirical + code review: the manuscript is read only to resolve table numbering, and prose is not examined.

2026-07-221

annotator-input-parity-check

kennethkhoocy/applied-micro-skills

Before designing, training, or auditing ANY model that replicates human-annotated labels, audit the annotation protocol's INPUT — the exact document/evidence the human labelers consulted — and give the model that same input. Use when: (1) designing a classifier/LLM extractor whose target is a hand-coded label set, (2) a label-replication model shows low recall concentrated in a label subset and the diagnosis on offer is "the label's information is not in the features", (3) reviewers propose construct splits (e.g. "designation vs record-evident"), adjudication sittings, or per-domain stop rules to explain residual disagreement with gold, (4) validating an extraction pipeline against labels transcribed from a source document. Symptom of the underlying failure: elaborate theory accumulates to explain why gold is "partially unpredictable" when the model was simply never shown the document the annotators read.

2026-07-221

asyncopenai-concurrency-httpx-pool

kennethkhoocy/applied-micro-skills

Raise real concurrency in asyncio LLM batch scorers built on the OpenAI SDK (AsyncOpenAI, including OpenAI-compatible providers like DeepSeek). Use when: (1) raising an asyncio.Semaphore above ~100 produces no throughput gain, (2) a batch pipeline saturates near 100 in-flight requests despite a larger semaphore, (3) planning a high-concurrency campaign against a provider with no hard rate limit (DeepSeek v4-flash tolerates 2000+ in flight). Root cause: AsyncOpenAI's default httpx pool caps max_connections at 100, silently bottlenecking any larger semaphore — you must pass a custom http_client with httpx.Limits sized to the semaphore.

2026-07-221

cite-placement

kennethkhoocy/applied-micro-skills

Place pre-screened literature citations into a LaTeX or Word manuscript, or restyle the citations already in one. Three modes: (1) inline placement — inline \cite{}/\citet{}/\citep{} with a compiled references.bib, for author-date journals (APA, MLA, Harvard, Chicago author-date, IEEE, Vancouver); (2) footnote placement — full formatted \footnote{} or OOXML footnotes for legal and notes styles (Bluebook, OSCOLA, Chicago, APA, McGill) with Id./supra short forms; (3) restyle — convert existing footnote citations from one style to another. This skill is manual-invoke ONLY — trigger ONLY when the user explicitly runs /cite-placement or explicitly names the "cite-placement" skill. Do NOT auto-trigger on general citation, footnote, or reference requests.

2026-07-221

download-gated-pdfs

kennethkhoocy/applied-micro-skills

Download the actual PDF binary from bot-gated sites (taxpolicycenter.org, urban.org, SSRN-hosted mirrors, think-tank/publisher sites) via the Wayback Machine id_ URL form. Use when: (1) curl/WebFetch of a .pdf URL returns HTML instead of a PDF even with a browser User-Agent, (2) pypdf fails with "invalid pdf header: b'<!DOC'" or "EOF marker not found" on a freshly downloaded file, (3) Firecrawl can parse the PDF to markdown but you need the original file on disk (e.g., filing a reference copy).

2026-07-221

name	pyfixest-cupy64-absorbed-regressors
description	pyfixest demeaner_backend="cupy64" (including its CPU fallback when cupy is absent) is NOT numerically identical to the default numba backend and does NOT drop fully-absorbed/collinear regressors the same way. Use when: (1) adding demeaner_backend="cupy64" to existing pf.feols/fepois calls changes the printed coefficient table, (2) a regression report suddenly gains rows with absurd estimates (e.g. coef 435.8, SE 7106) for controls absorbed by the fixed effects, (3) diffing outputs before/after a backend change, or (4) anything parses a pyfixest text report by line position.
author	Claude Code
version	1.0.0
date	"2026-07-18T00:00:00.000Z"

pyfixest cupy64 backend: absorbed regressors survive, reports change shape

Problem

Context / Trigger Conditions

Solution

Treat a backend change as a POTENTIALLY OUTPUT-CHANGING edit: diff the report and expect schema changes, not byte equality. Verify the coefficients of interest at ~4-decimal precision instead.
Never interpret retained absorbed-regressor rows as estimates; check within-FE variation before reading nuisance coefficients.
Never parse pyfixest text reports by line position; anchor on the variable name of the coefficient you need.
When byte-stable reports matter (regression-tested pipelines), pin the backend consistently everywhere rather than mixing backends across runs.

Verification

Notes

Applies even with no GPU: the fail-open CPU fallback shows the same retention behavior, so "cupy isn't installed" does not make the kwarg inert.
The headline inference (identified coefficients, clustered SEs) agrees to reporting precision; this is a report-shape/nuisance-row issue, not a correctness issue for identified estimates.