Implementation:Tensorflow Tfjs Container Test

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

Overview

This test suite validates the Container class, which is the foundation of the Functional API model in TensorFlow.js Layers. Containers manage the computational graph of layers, tracking inputs, outputs, and the topology of connected layers. The tests cover container construction from configuration (including deserialization from Python Keras configs), layer connectivity, loss calculation, source input resolution, and model disposal/memory management.

Code Reference

Source Location: tfjs-layers/src/engine/container_test.ts (695 lines)

Repository: GitHub

Test Describe Blocks

Container.fromConfig - Creating containers from serialized config (minimal, simple network, multi-node, multi-output)
Container - Layer connectivity, weight management, naming, output shape, stateful properties, call hooks, nested models
Container.calculateLosses - Kernel and bias regularization loss computation
getSourceInputs() - Tracing back through the graph to find source input tensors
LayersModel-dispose - Memory management and tensor disposal when freeing models

I/O Contract

Inputs to tests:

Serialized model configurations matching Python Keras output (JSON with layer configs, inbound nodes, input/output layers)
Layer instances connected via SymbolicTensors
Regularizer configurations (L1L2)

Expected outputs/assertions:

Containers reconstruct correct topology from config (layer names, connections, shapes)
Output shapes are correctly computed from the graph
Weight counts match expected values
Regularization losses are computed correctly
Model disposal frees all weight tensors and reduces memory tensor count

Usage Example

describeMathCPUAndGPU('Container.fromConfig', () => {
  it('creates a minimal Container from simplest config', () => {
    const config = {
      name: 'test',
      layers: [] as any[],
      inputLayers: [] as any[],
      outputLayers: [] as any[]
    };
    const container =
        Container.fromConfig(ContainerForTest, config) as Container;
    expect(container.name).toEqual('test');
  });
});

Test Coverage Summary

Category	Count	Details
fromConfig	3+	Minimal, simple network, multi-node topologies
Container Properties	10+	Weights, naming, output shape, stateful, call hooks
Loss Calculation	3+	Kernel regularization, bias regularization, layer losses
Source Inputs	2+	Tracing through graph to source inputs
Disposal	3+	Memory cleanup, tensor count verification
Test Environment	CPU and GPU	`describeMathCPUAndGPU`

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment