Art of DS

Appearance

Palette
Theme
Unlock $20 one time
Sign in, or create an account

Sign in to Art of DSCreate your Art of DS account

We email you a link and a six-digit code. Either one signs you in; there is no password.Enter your email. We send a link and a six-digit code to confirm it — no password to remember. The account is free.

  • You come back to this page, right where you were.
  • Already have an account? The same email simply signs you in.

By continuing you accept the Terms and the Privacy policy.

Your account

Free accountLifetime access

Knowledge map

Every page names what to read before it. Here is the whole library at once: a dot for each page, its topic down the side, and how far it stands from a starting point across. Pick a dot to see the way to it.

Pages
126
Links between them
121
The longest way in
13 pages
The board: every page, by topic and by distance from a starting pointStart12345678Steps from a starting point →Mathematics4Statistics15Metrics & losses21Optimization8Machine learning30Deep learning16Computer vision13NLP17Tooling2619721031145812GradientExpectation and varianceEigenvectors and eigenvaluesHypothesis testingBayes' theoremA/B testingTime series decompositionMSEConfusion matrixMAEIntersection over unionDecision treeKNNPCAK-meansTrain-test splitImage classificationTokenizationNatural language processingJupyter NotebookMachine learning workflowEntropyCorrelationOutliersStationarityConfidence intervalConfoundingMaximum likelihoodMultiple comparisonsAccuracyHuber lossMAPEMSLEPrecisionR²RecallRMSESpecificityBLEUGradient descentMissing dataClass imbalanceLinear regressionDistance-weighted KNNManifold learningSVMBackpropagationBag of wordsLanguage modelNamed entity recognitionWord embeddingsAutocorrelationBootstrapCentral limit theoremCross-entropyMean average precisionF1-scoreROC AUCPerplexityBFGSConjugate gradientLearning rateSGDRegularizationFeature engineeringLogistic regressionOverfittingPolynomial regressionPerceptronRecurrent neural networkGenerative adversarial networkDecodingPromptingTF-IDFAttentionSentence embeddingsGeneralized linear modelsModel calibrationAdamMini-batch gradient descentEarly stoppingGaussian discriminant analysisLasso regressionRidge regressionBaggingBoostingBias-variance tradeoffCross-validationDropoutConvolutionSigmoidData augmentationPointNetDiffusion modelSequence-to-sequenceHallucinationInstruction tuningRetrieval-augmented generationFréchet Inception DistanceGradient boostingBayesian linear regressionElastic NetRandom forestData leakageHyperparameter tuningBatch normalizationDepthwise separable convolutionPoolingReceptive fieldReLUSoftmaxTanhTransposed convolutionConvolutional neural networkFeature importanceResidual connectionLeaky ReLUObject detectionSelf-supervised learningSemantic segmentationTransfer learningInstance segmentationTransformerVision transformerCLIP
  • Free
  • With access
  • Read first, in this order
  • Builds on it
  • Read
The way to BERT 12 pages to read first

NLP · difficulty 4/5

BERT

A transformer encoder pretrained by hiding 15% of the tokens and predicting them from both sides at once, then fine-tuned on a task with one extra output layer and no architectural surgery.

The way here

12 pages to read first · 5 of them free

  1. 1MSEMetrics & losses · 1/5
  2. 2Linear regressionMachine learning · 2/5
  3. 3PerceptronDeep learning · 2/5
  4. 4ConvolutionDeep learning · 3/5with access
  5. 5Convolutional neural networkComputer vision · 3/5with accessafter 3. Perceptron
  6. 6GradientMathematics · 2/5
  7. 7Gradient descentOptimization · 2/5
  8. 8Residual connectionDeep learning · 4/5with accessafter 5. Convolutional neural network
  9. 9TokenizationNLP · 2/5with access
  10. 10Word embeddingsNLP · 3/5with access
  11. 11AttentionNLP · 4/5with access
  12. 12TransformerNLP · 4/5with accessafter 8. Residual connection
  13. BERTThe page you pickedafter 11. Attention

What builds on it

Nothing builds on it yet.

Have a promo code?

A code that takes 100% off unlocks everything at once. A smaller one takes its share off the price in the payment window: enter the same code there.

Write to support

A question, a problem with access, a refund? Write to us and we answer by email.

What is it about?

Your address is used only to answer you.

Message sent

Thank you. We will reply to the address you gave, by email.

Or write to us directly

support@artofds.com

Message sent Thank you. We will reply to the address you gave, by email.

Check the email address: we need it to reply.

That is our own address. Enter yours, so the reply reaches you.

This address cannot receive email. Check it for a typo.

Write a little more: at least 10 characters.

Too many messages in a short time. Try again in a few minutes.

The message could not be sent. Please write to us directly: support@artofds.com