mirror of
https://github.com/MLSysBook/TinyTorch.git
synced 2026-03-12 03:03:37 -05:00
06b35c34bd
Major fixes for complete training pipeline functionality: Core Components Fixed: - Parameter class: Now wraps Variables with requires_grad=True for proper gradient tracking - Variable.sum(): Essential for scalar loss computation from multi-element tensors - Gradient handling: Fixed memoryview issues in autograd and activations - Tensor indexing: Added __getitem__ support for weight inspection Training Results: - XOR learning: 100% accuracy (4/4) - network successfully learns XOR function - Linear regression: Weight=1.991 (target=2.0), Bias=0.980 (target=1.0) - Integration tests: 21/22 passing (95.5% success rate) - Module tests: All individual modules passing - General functionality: 4/5 tests passing with core training working Technical Details: - Fixed gradient data access patterns throughout activations.py - Added safe memoryview handling in Variable.backward() - Implemented proper Parameter-Variable delegation - Added Tensor subscripting for debugging access(https://claude.ai/code)
122 lines
3.8 KiB
Python
122 lines
3.8 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Test that the fixed training pipeline works correctly.
|
|
"""
|
|
|
|
import numpy as np
|
|
import sys
|
|
import os
|
|
|
|
# Add the tinytorch package to path
|
|
sys.path.insert(0, os.path.abspath('.'))
|
|
|
|
from tinytorch.core.tensor import Tensor
|
|
from tinytorch.core.layers import Parameter, Linear
|
|
from tinytorch.core.autograd import Variable
|
|
from tinytorch.core.optimizers import SGD
|
|
from tinytorch.core.training import MeanSquaredError
|
|
|
|
def test_linear_regression_extended():
|
|
"""Test linear regression with more epochs."""
|
|
print("🔬 Testing Extended Linear Regression...")
|
|
|
|
# Generate simple data: y = 2x + 1
|
|
np.random.seed(42)
|
|
X = np.random.randn(50, 1).astype(np.float32)
|
|
y = 2 * X + 1 + 0.01 * np.random.randn(50, 1)
|
|
|
|
print(f"Training on {len(X)} samples")
|
|
print(f"Target function: y = 2x + 1")
|
|
|
|
# Create model
|
|
model = Linear(1, 1)
|
|
optimizer = SGD(model.parameters(), learning_rate=0.1)
|
|
loss_fn = MeanSquaredError()
|
|
|
|
# Training loop with more epochs
|
|
losses = []
|
|
for epoch in range(150):
|
|
# Zero gradients
|
|
optimizer.zero_grad()
|
|
|
|
# Forward pass
|
|
X_var = Variable(X, requires_grad=False)
|
|
predictions = model(X_var)
|
|
|
|
# Convert y to Variable
|
|
y_var = Variable(y, requires_grad=False)
|
|
|
|
# Compute loss
|
|
loss = loss_fn(predictions, y_var)
|
|
losses.append(loss.data.data)
|
|
|
|
if epoch % 50 == 0:
|
|
print(f"Epoch {epoch:3d}: Loss = {loss.data.data:.6f}")
|
|
|
|
# Backward pass
|
|
loss.backward()
|
|
|
|
# Update parameters
|
|
optimizer.step()
|
|
|
|
# Check final parameters
|
|
params = list(model.parameters())
|
|
final_weight = params[0].data.data[0, 0]
|
|
final_bias = params[1].data.data[0]
|
|
|
|
print(f"\nFinal Results:")
|
|
print(f" Weight: {final_weight:.3f} (target: 2.0)")
|
|
print(f" Bias: {final_bias:.3f} (target: 1.0)")
|
|
print(f" Final Loss: {losses[-1]:.6f}")
|
|
|
|
# Check if parameters are close to target
|
|
weight_error = abs(final_weight - 2.0)
|
|
bias_error = abs(final_bias - 1.0)
|
|
|
|
print(f"\nErrors:")
|
|
print(f" Weight error: {weight_error:.3f}")
|
|
print(f" Bias error: {bias_error:.3f}")
|
|
|
|
# Success criteria: errors should be small
|
|
weight_ok = weight_error < 0.5 # Within 0.5 of target
|
|
bias_ok = bias_error < 0.5 # Within 0.5 of target
|
|
loss_decreased = losses[-1] < losses[0] * 0.1 # Loss reduced by 90%
|
|
|
|
success = weight_ok and bias_ok and loss_decreased
|
|
|
|
if success:
|
|
print("✅ Linear regression training successful!")
|
|
print(f" Parameters learned correctly (errors < 0.5)")
|
|
print(f" Loss reduced by {(1 - losses[-1]/losses[0])*100:.1f}%")
|
|
else:
|
|
print("❌ Linear regression training failed:")
|
|
if not weight_ok:
|
|
print(f" Weight error too large: {weight_error:.3f}")
|
|
if not bias_ok:
|
|
print(f" Bias error too large: {bias_error:.3f}")
|
|
if not loss_decreased:
|
|
print(f" Loss did not decrease enough: {losses[0]:.3f} → {losses[-1]:.3f}")
|
|
|
|
return success
|
|
|
|
if __name__ == "__main__":
|
|
print("🔥 TinyTorch Training Pipeline Test")
|
|
print("=" * 50)
|
|
|
|
try:
|
|
success = test_linear_regression_extended()
|
|
|
|
if success:
|
|
print("\n🎉 Training pipeline is working correctly!")
|
|
print("✅ Parameter class creates Variables with gradient tracking")
|
|
print("✅ Variable.sum() enables scalar loss computation")
|
|
print("✅ Gradient flow works from loss back to parameters")
|
|
print("✅ Optimizer updates parameters correctly")
|
|
print("✅ Linear regression learns target function")
|
|
else:
|
|
print("\n❌ Training pipeline has issues")
|
|
|
|
except Exception as e:
|
|
print(f"\n❌ Error: {e}")
|
|
import traceback
|
|
traceback.print_exc() |