Skip to learning content

DATA SCIENCE PYTHON PLAYGROUND

Machine Learning · Learn / Refresh

← Choose a deckQUESTIONS · MODELS · EVIDENCE
DECK 4

Neural Networks

Shared network ideas, then a regression or classification route.

01 / Shared network concepts

N01Core · teaching
From inputs to network outputs: Input features, Hidden units, OutputInput featuresHidden unitsOutput

From inputs to network outputs

Understand layers, units and task-shaped outputs.

12–18 min · 3 practices
N02Core · teaching
Learning weights by reducing loss: Current weights, Training loss, Update weightsCurrent weightsTraining lossUpdate weights

Learning weights by reducing loss

Separate optimisation from generalisation.

12–18 min · 3 practices
N05Core · teaching
Convergence, early stopping and validation: Training loss, Validation gap, Stop in timeTraining lossValidation gapStop in time

Convergence, early stopping and validation

Distinguish internal stopping, outer CV and final testing.

18–25 min · 4 practices
N-R1Core · review
Shared network retrieval: Inputs and units, Loss and convergence, Class or valueInputs and unitsLoss and convergenceClass or value

Shared network retrieval

Retrieve earlier concepts before combining them.

15–20 min · 3 practices

02 / Task workflows

N03Core · teaching
Neural classification workflow: Scaled X, Class outputs, Macro F1Scaled XClass outputsMacro F1

Neural classification workflow

Apply class outputs and classification evidence to a small network.

25–40 min · 4 practices
N04Core · teaching
Neural regression workflow: Scaled X, One output, Original unitsScaled XOne outputOriginal units

Neural regression workflow

Keep feature scaling and target transformation distinct.

18–25 min · 4 practices

03 / Go Further

N-K2Core · checkpoint
Neural classification checkpoint: Penguin X, Tune network, Class errorsPenguin XTune networkClass errors

Neural classification checkpoint

Complete the Penguin neural classifier workflow with a reference, width selection, class diagnostics and final metrics.

25–35 min · 1 practices
N-K1Core · checkpoint
Neural regression checkpoint: Wine X, Tune network, Original RMSEWine XTune networkOriginalRMSE

Neural regression checkpoint

Complete the Wine600 neural regression workflow with feature/target scaling, width selection, convergence evidence and original-unit errors.

25–35 min · 1 practices
N06Go Further · teaching
Capacity and regularisation: More capacity, Regularisation, ValidationCapacityPenaltyTune both

Capacity and regularisation

Interpret controlled width and alpha experiments.

12–18 min · 3 practices