mirror of
https://github.com/harvard-edge/cs249r_book.git
synced 2026-08-03 12:14:39 -05:00
Vijay Janapa Reddi
d30257577c
Replace np.random.randn/rand/seed with np.random.default_rng(7) across all 93 source modules, tests, and milestones for reproducible, isolated random state.
118 lines
3.4 KiB
Python
118 lines
3.4 KiB
Python
"""
|
|
Tests for Module 04: Linear/Networks
|
|
"""
|
|
|
|
import pytest
|
|
import numpy as np
|
|
rng = np.random.default_rng(7)
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
# Add project root to path
|
|
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
|
|
|
|
|
|
class TestLinearExports:
|
|
"""Test that Linear layer is properly exported."""
|
|
|
|
def test_dense_import(self):
|
|
"""Test Linear can be imported from correct location."""
|
|
from tinytorch.core.layers import Linear
|
|
assert Linear is not None
|
|
|
|
def test_dense_creation(self):
|
|
"""Test Linear layer can be created."""
|
|
from tinytorch.core.layers import Linear
|
|
layer = Linear(10, 5)
|
|
assert layer.weight.shape == (10, 5)
|
|
|
|
|
|
class TestLinearForward:
|
|
"""Test Linear layer forward pass."""
|
|
|
|
def test_forward_shape(self):
|
|
"""Test output shape is correct."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer = Linear(10, 5)
|
|
x = Tensor(rng.standard_normal((32, 10)))
|
|
output = layer(x)
|
|
|
|
assert output.shape == (32, 5)
|
|
|
|
def test_forward_with_bias(self):
|
|
"""Test forward pass with bias."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer = Linear(10, 5, bias=True)
|
|
x = Tensor(np.zeros((1, 10)))
|
|
output = layer(x)
|
|
|
|
# With zero input, output should equal bias
|
|
assert np.allclose(output.data, layer.bias.data)
|
|
|
|
def test_forward_without_bias(self):
|
|
"""Test forward pass without bias."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer = Linear(10, 5, bias=False)
|
|
x = Tensor(np.zeros((1, 10)))
|
|
output = layer(x)
|
|
|
|
# With zero input and no bias, output should be zero
|
|
assert np.allclose(output.data, 0)
|
|
|
|
|
|
class TestLinearIntegration:
|
|
"""Test Linear layer integration with other modules."""
|
|
|
|
def test_dense_with_tensor(self):
|
|
"""Test Linear works with Tensor (Module 02)."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer = Linear(10, 5)
|
|
|
|
# Weights and bias should be Tensors
|
|
assert isinstance(layer.weight, Tensor)
|
|
if layer.bias is not None:
|
|
assert isinstance(layer.bias, Tensor)
|
|
|
|
def test_dense_with_activations(self):
|
|
"""Test Linear works with activations (Module 03)."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.activations import ReLU, Sigmoid
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer = Linear(10, 5)
|
|
relu = ReLU()
|
|
sigmoid = Sigmoid()
|
|
|
|
x = Tensor(rng.standard_normal((16, 10)))
|
|
h = layer(x)
|
|
h_relu = relu(h)
|
|
h_sigmoid = sigmoid(h)
|
|
|
|
assert h_relu.shape == h.shape
|
|
assert h_sigmoid.shape == h.shape
|
|
assert np.all(h_sigmoid.data >= 0) and np.all(h_sigmoid.data <= 1)
|
|
|
|
def test_dense_chain(self):
|
|
"""Test chaining multiple Linear layers."""
|
|
from tinytorch.core.layers import Linear
|
|
from tinytorch.core.tensor import Tensor
|
|
|
|
layer1 = Linear(784, 128)
|
|
layer2 = Linear(128, 64)
|
|
layer3 = Linear(64, 10)
|
|
|
|
x = Tensor(rng.standard_normal((32, 784)))
|
|
h1 = layer1(x)
|
|
h2 = layer2(h1)
|
|
output = layer3(h2)
|
|
|
|
assert output.shape == (32, 10)
|