#machine-learning

Classifier-evaluation maths as an API, computed locally and deterministically. The confusion endpoint turns the four cells of a binary confusion matrix — true and false positives and negatives — into the full metric suite: accuracy, precision, recall (sensitivity), specificity, the F1 score, the Matthews correlation coefficient (robust to class imbalance), balanced accuracy, negative predictive value, the false-positive and false-negative rates and the prevalence. The diagnostic endpoint applies Bayes' theorem to a medical or screening test: from its sensitivity, specificity and the prevalence (pre-test probability) it gives the positive and negative predictive values, the positive and negative likelihood ratios and the diagnostic odds ratio. The fbeta endpoint computes the Fβ score from precision and recall (or from the raw counts) for any β — β = 1 is F1, larger β weights recall, smaller β weights precision. Metrics whose denominator is zero are returned as null rather than erroring. Everything is computed locally and deterministically, so it is instant and private. Ideal for machine-learning, data-science, medical-testing and analytics app developers, model-evaluation and screening tools, and statistics education. Pure local computation — no key, no third-party service, instant. Live, nothing stored. 3 endpoints. This is classifier evaluation; for descriptive statistics and regression use a statistics API and for hypothesis tests an inference API.

api.oanor.com/classifier-api

The Hugging Face Hub as an API — the central, open registry of machine-learning models and datasets that powers much of the modern AI ecosystem. This API wraps the public huggingface.co Hub into clean JSON. /v1/models searches the Hub's models and lets you filter by task (pipeline_tag — e.g. text-generation, text-to-image, image-classification, automatic-speech-recognition, sentence-similarity) and by library (transformers, diffusers, sentence-transformers, …), sorted by downloads, likes, last-modified, created or trending score — each model returned with its id, author, task, library, download and like counts, license, tags and timestamps. /v1/model?id=google-bert/bert-base-uncased returns a single model's full metadata. /v1/datasets searches ML datasets the same way, and /v1/dataset?id=ILSVRC/imagenet-1k returns a single dataset's metadata. Ids are in org/name form (take them from the search endpoints). Ideal for ML and MLOps tooling, model-discovery and comparison sites, AI leaderboards and dashboards, and AI assistants that recommend models. Data comes from the public Hugging Face Hub (free to use). This is the AI/ML model and dataset hub — distinct from software-package registries (npm, PyPI, Maven, NuGet) and academic paper indexes (arXiv).

api.oanor.com/huggingface-api

Detect human faces in an image and analyse each one with on-device machine learning: get the bounding box and a detection confidence, an estimated age, the predicted gender with its probability, and the dominant facial expression together with the full per-expression breakdown (neutral, happy, sad, angry, fearful, disgusted and surprised). A lightweight count endpoint returns just the number of faces and their boxes for fast gating. Supply an image by public URL, base64 or a raw binary request body; only public http/https URLs are accepted and private or internal hosts are blocked, and large images are downscaled automatically. Runs locally on TensorFlow (face-api) — no third-party upstream and no per-image cloud cost — with warm models that keep inference fast. Ideal for photo and avatar apps, audience analytics, smart cameras, auto-cropping and accessibility.

api.oanor.com/facedetect-api

Moderate images automatically with on-device machine learning. Classify any image across five categories — neutral, drawing, sexy, porn and hentai — and receive per-class probabilities, the top class, a combined NSFW score and a clear verdict (safe, questionable or nsfw). A simpler check endpoint returns a single safe/unsafe decision against a threshold you choose, ideal for upload gates and user-generated-content pipelines. Supply an image by public URL, base64 or a raw binary request body; only public http/https URLs are accepted and private or internal hosts are blocked, and large images are downscaled automatically. Runs locally on TensorFlow (NSFWJS / MobileNetV2) — no third-party upstream and no per-image cloud cost — with a warm model that keeps inference fast. Ideal for community platforms, marketplaces, dating and chat apps, and any service that accepts user images.

api.oanor.com/nsfw-api