Solutions · Research and teaching

Explanations you can cite, rerun and teach.

For academic groups, research labs and public bodies. Exact where it can be, seeded where it samples, and open about its methods: results another group can reproduce, and a model anatomy students can walk through weight by weight.

Free plan
10 models500 explanations, 20 analyses and 5,000 logged predictions a month
Client
Apache 2.0an open Python client; every call is also plain REST
Worked examples
6on public datasets, each number read back from the API
Open notebooks
5the Titanic walkthrough, from upload to drift, ready for Colab

What you get

Built to be checked.

The properties a reviewer, a co-author or a student will ask about.

Reproducible

The same request, the same numbers

KernelSHAP enumerates every feature coalition for models with up to 12 inputs, so the result is exact for the reference sample. Sampling methods take a seed.

On the record

Every result knows how it was made

An explanation is stored with its method, options, seed, model version and reference sample. An analysis is stored with its row counts and model calls.

Exportable

JSON in, JSON out

Structures, attributions, behaviour curves and drift reports come back as JSON, ready for your own figures and tables.

Teachable

A model students can walk through

Model anatomy draws every layer, split and weight, and a trace follows one prediction through them.

Honest

Limits stated with the result

Permutation sampling reports a margin of error for each attribution. LIME reports how well its local fit holds. Non-monotonic effects are named as such.

Private

Unpublished models stay yours

Self-host the API, explain through your own endpoint, or upload shapes only.

Methods

Standard methods, named and cited.

Kaimei implements published methods and says which one produced each number.

MethodWhat Kaimei computesReference
KernelSHAPShapley values against a reference sample; exact by enumeration up to 12 inputs.Lundberg & Lee, 2017
Permutation ShapleyMonte Carlo Shapley values over random feature orderings, with a standard error per input.Štrumbelj & Kononenko, 2014
LIMEA weighted local linear surrogate, for tables and image patches, with its fit reported.Ribeiro, Singh & Guestrin, 2016
Grad-CAMClass evidence on the last convolutional feature map of an uploaded network.Selvaraju et al., 2017
OcclusionThe change in the class score as patches of the image are hidden.Zeiler & Fergus, 2014
Permutation importanceHow far the model's output moves when one input is shuffled across the reference rows.Breiman, 2001
Partial dependence and ICEThe average and per-row response as one input is varied, with a monotonicity score.Friedman, 2001; Goldstein et al., 2015
Interaction strengthThe share of two inputs' joint effect that is not additive.
Population stability indexDrift of each input, and of the output, between reference rows and logged predictions.

Worked examples

Six public datasets, six models, one way of working.

ExampleDataModel
Banking and lendingRead the exampleYeh, I.-C. and Lien, C.-H. (2009). The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients. Expert Systems with Applications 36(2). Sourcecredit-default-forest
scikit-learn random forest
Capital marketsRead the exampleKenneth R. French Data Library, 49 Industry Portfolios, monthly value-weighted returns. Sourcesector-rotation
scikit-learn gradient boosting
InsuranceRead the exampleDutang, C. and Charpentier, A. CASdatasets: freMTPL2freq, French motor third-party liability claims. Sourcemotor-frequency
scikit-learn gradient boosting
HealthcareRead the exampleStrack, B. et al. (2014). Impact of HbA1c measurement on hospital readmission rates: analysis of 70,000 clinical database patient records. BioMed Research International. Sourcereadmission-30d
scikit-learn gradient boosting
TechnologyRead the exampleIBM sample data: Telco Customer Churn. Sourcechurn-risk-gbm
scikit-learn gradient boosting
Energy and industryRead the exampleSalam, A. and El Hibaoui, A. (2018). Comparison of machine learning algorithms for the power consumption prediction: case study of Tetouan city. IRSEC. Sourceload-forecast
scikit-learn neural network
  1. Public dataEach example starts from a dataset anyone can download, cited above.
  2. An ordinary modelTrained with scikit-learn in the usual way, on a held-out split, with a fixed seed.
  3. One uploadSent to Kaimei with client.models.upload, with reference rows drawn from the training data.
  4. Read backEvery attribution, insight and drift figure on these pages is what the API returned. Accuracy figures are computed on the held-out split.

Teaching

Three places to start a class.

Walk a forest, tree by tree

The live model anatomy draws a real random forest and lights up the path one row takes through it, split by split.

Model anatomy

Five notebooks, start to finish

The Titanic walkthrough is the teaching example: upload, explain, analyse, monitor and version a model, with every output saved.

The notebooks

What an explanation is not

A short guide to reading attributions without over-reading them: not causation, not a fairness test, not a check of the data.

Reading an explanation

Who it is for

Groups, classrooms and public bodies.

01

Research groups

Exact and seeded attributions for a paper's figures, with the options that reproduce them stored next to the result.

02

Lecturers and students

A free plan that fits a course, an open client, and a model anatomy that makes a trained network something you can point at.

03

Public bodies and reviewers

A plain-English account of what a model relies on, and a record of each decision and its explanation, kept under your own retention rules.

Coming soon

Next on the roadmap.

What we are building now, and what to use until each one ships.

  • Coming soonText and language models

    Token and passage attributions for text classifiers and language models, and for the score of a chosen answer from a generative model.

  • Coming soonSequences, audio and video

    Inputs that arrive step by step or as a signal, attributed per time step, frequency band or frame.

  • Coming soonFairness metrics

    Group fairness and disparate-impact measures as their own analysis, stored with its seed and options like every other result.

  • Coming soonIntegrated Gradients and DeepLIFT

    Gradient-based attributions computed by Kaimei for uploaded networks. Until then, store the ones you compute beside the built-in methods.

  • Coming soonAdversarial attacks

    Gradient attacks such as FGSM and PGD beside the robustness analysis, which measures decision flips under random noise today.

  • Coming soonAn R client

    The same calls from R, for statisticians and for teaching. Until then, the REST API works from any language.

See it on your own model.

Start on the free plan, or write to us about a course or a research project.