Implementation:Tensorflow Tfjs Topology Test

Knowledge Sources	Tensorflow_Tfjs
Domains	Testing, Layers_API
Last Updated	2026-02-10 06:00 GMT

Overview

This test suite validates the core topology primitives of the TensorFlow.js Layers engine: InputSpec, Node, SymbolicTensor, and Layer. These are the building blocks that define how layers connect and data flows through a neural network graph. The tests cover input specification validation, node creation and serialization, layer construction (naming, weight management, trainability, state), layer apply behavior with both symbolic and concrete tensors, input compatibility checking, and weight loading from named tensor maps.

Code Reference

Source Location: tfjs-layers/src/engine/topology_test.ts (1181 lines)

Repository: GitHub

Test Describe Blocks

InputSpec - Default values, ndim from shape, axes specification
Node - Object initialization, serializable node config generation
Layer - Comprehensive layer tests including:
- Constructor (naming, auto-incrementing, dtype, initial weight)
- apply() with SymbolicTensors and concrete tensors
- Weight management (trainable/non-trainable, setting weights, counting params)
- Input compatibility checking (ndim, dtype, axes constraints)
- Stateful layers (resetStates)
- Masking support (supportsMasking, computeMask)
- Config serialization (getConfig)
Layer-dispose - Disposing layers and verifying tensor cleanup
loadWeightsFromNamedTensorMap - Loading weights from key-value tensor maps with strict/non-strict modes

I/O Contract

Inputs to tests:

Layer configurations (name, dtype, inputShape, batchInputShape, trainable)
SymbolicTensors with known shapes and dtypes
Concrete tensors (zeros, ones, random) for apply() calls
Named tensor maps for weight loading

Expected outputs/assertions:

InputSpec properties match constructor args
Nodes correctly track inbound/outbound layers and tensor indices
Layer names auto-increment and respect name scopes
apply() produces correct output shapes for both symbolic and concrete tensors
Input compatibility errors thrown for mismatched ndim, dtype, or axes
Weight loading correctly matches tensors by name, with strict mode errors for missing weights

Usage Example

describe('InputSpec', () => {
  it('initializes with expected default values.', () => {
    const inputSpec = new InputSpec({});
    expect(inputSpec.dtype).toBeUndefined();
    expect(inputSpec.shape).toBeUndefined();
    expect(inputSpec.ndim).toBeUndefined();
    expect(inputSpec.maxNDim).toBeUndefined();
    expect(inputSpec.minNDim).toBeUndefined();
    expect(inputSpec.axes).toEqual({});
  });
});

describe('Node', () => {
  it('initializes object as expected.', () => {
    const outboundLayer = new LayerForTest({name: 'outboundLayer'});
    const inboundLayers = [new LayerForTest({name: 'inboundLayer'})];
    const node = new Node({
      outboundLayer, inboundLayers,
      nodeIndices: [0], tensorIndices: [0],
      inputTensors: [new tfl.SymbolicTensor('float32', [1], null, [], {})],
      outputTensors: [new tfl.SymbolicTensor('float32', [2, 2], null, [], {})],
      inputMasks: [zeros([1])], outputMasks: [zeros([1])],
      inputShapes: [[1]], outputShapes: [[1], [1]]
    }, {});
    expect(node.outboundLayer).toEqual(outboundLayer);
  });
});

Test Coverage Summary

Category	Count	Details
InputSpec	3	Default values, ndim inference, axes
Node	2+	Initialization, serialization
Layer Construction	15+	Naming, dtype, weights, trainability
Layer.apply()	10+	Symbolic and concrete tensor modes
Input Compatibility	5+	ndim, dtype, axes validation
Layer Disposal	5+	Tensor cleanup, shared weight handling
Weight Loading	5+	Named tensor map, strict mode
Test Environment	Mixed	`describeMathCPU`, `describeMathCPUAndGPU`

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment