SShortSingh.
Back to feed

Time-weighted EWM in pandas smooths irregular stock sentiment without overnight lag

0
·1 views

A DEV Community walkthrough shows how pandas' ewm with halflife as a time delta weights sentiment snippets by elapsed clock time rather than row count. This suits irregularly timed news, research notes, and forum posts about a single stock, which arrive in dense bursts during trading and sparsely overnight. The tutorial uses synthetic scores and placeholder labels, and it demonstrates weight checks, a per-stock groupby, and an effective-sample-size metric to flag thin data. It also warns that an EWM only updates when new rows arrive, so a stale value can persist through long silences. A decayed-mean function is introduced to evaluate the average at any current time, optionally drifting toward a neutral prior.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Brevia 2.x adds local meeting transcription, source-linked notes, and MCP support

Brevia is an open-source app for recording conversations, transcribing audio, and keeping meeting notes tied to the original speech, running on macOS, Windows, and Linux with companion iPhone and Android apps. Its minutes link conclusions to transcript passages and replayable timestamps, and AI suggestions for decisions and action items are accepted individually by the user. The app supports transcription in over 30 languages, speaker identification, glossary management for unfamiliar terms, and exports to several formats. Recordings stay on the user's devices by default, with offline processing available once required models are downloaded. Desktop installers range from roughly 38 MB to 180 MB, excluding model weights, and the app now runs on Tauri/Rust rather than Electron and Python.

0
ProgrammingDEV Community ·

Ankita AI adds durable tasks so desktop assistant resumes after crashes without repeating completed actions

Ankita AI, a desktop assistant built by Krish Verma that writes files, edits code, and sends messages, previously risked duplicating effects when a crash forced it to restart a job. The author introduced durable tasks in a 242-line AssistantTaskStore, where each request becomes a checkpointed record saved atomically with size and step limits. Each action is tracked through phases such as requested, started, settled, uncertain, and not_run, and a hash of the tool name and sorted arguments blocks any repeat of an action that already completed. Tasks interrupted by a dead process are recovered only when the owning process is confirmed gone, and uncertain steps require owner review before further changes proceed. In a restart test with a real process killed mid-task, the completed write was blocked from repeating while the remaining work finished.

0
ProgrammingDEV Community ·

Katpolis: A Browser-Based Cat City Whose Adoptions Fund a Real Animal Shelter

Katpolis is an interactive, browser-based digital cat metropolis where each virtual cat adopted contributes to building and running a physical animal shelter. The developers chose React 19, Vite, and Cloudflare edge delivery over engines like Unity to keep load times short and avoid heavy downloads. Each cat's appearance is generated procedurally from a deterministic genetic seed, which reduced the asset payload from tens of megabytes to a few dozen kilobytes. The team also handled SEO for its single-page app by dynamically managing document titles, canonical tags, and OpenGraph metadata.

0
ProgrammingDEV Community ·

Castmagic vs Descript in 2026: Asset Extraction Versus Timeline Editing

A DEV Community piece compares Castmagic and Descript for 2026 content operations, judging both on throughput, latency, API reliability, and total cost of ownership. Castmagic is built to turn raw audio like podcasts and webinars into marketing assets such as show notes, social posts, and newsletter drafts using LLM prompt chains. Descript functions as a non-linear, transcript-driven editor suited to trimming audio, cleaning sound, and composing video. The article's comparison covers pricing at scale, learning curve, API and webhook options, ecosystem integrations, and security certifications. It concludes that the right choice depends on whether a team's bottleneck is content repurposing or post-production finishing.