How indranramesh/dl-from-scratch works
This repository is a small learning exercise demonstrating binary classification with a single neuron. Its main workflow generates synthetic examples, computes a linear score, applies sigmoid, and evaluates binary cross-entropy; a NumPy script also contains a manual gradient-update loop. A separate PyTorch script demonstrates the same task with automatic differentiation. The excerpts show both implementations, but do not establish that one invokes the other. The NumPy loop does not recompute predictions after updating its parameters.
Main entry points: train_torch.py Binary Cross-Entropy · forward_pass.py Manual Gradients
Components 20
NumPy Walkthrough
- NumPy Script
forward_pass.py - Synthetic Data
forward_pass.py - Binary Labels
forward_pass.py - Initial Parameters
forward_pass.py - Weighted Sum
forward_pass.py - Sigmoid
forward_pass.py - Cross-Entropy Loss
forward_pass.py - Manual Gradients
forward_pass.py - Parameter Update
forward_pass.py
PyTorch Walkthrough
- PyTorch Script
train_torch.py - Synthetic Data
train_torch.py - Input Tensors
train_torch.py - Tracked Parameters
train_torch.py - Linear Score
train_torch.py - Sigmoid
train_torch.py - Binary Cross-Entropy
train_torch.py - Autograd Backward
train_torch.py - Parameter Update
train_torch.py - Training Report
train_torch.py
Other components
- Learner
Connections 25
- Learner → NumPy Script: runs
- NumPy Script → Synthetic Data: generates
forward_pass.py - Synthetic Data → Binary Labels: derives
forward_pass.py - Synthetic Data → Weighted Sum: multiplies
forward_pass.py - Initial Parameters → Weighted Sum: weights
forward_pass.py - Weighted Sum → Sigmoid: feeds
forward_pass.py - Sigmoid → Cross-Entropy Loss: scores against
forward_pass.py - Binary Labels → Cross-Entropy Loss: labels
forward_pass.py - Cross-Entropy Loss → Manual Gradients: reports loss
forward_pass.py - Sigmoid → Manual Gradients: supplies predictions
forward_pass.py - Binary Labels → Manual Gradients: supplies targets
forward_pass.py - Manual Gradients → Parameter Update: updates
forward_pass.py - Learner → PyTorch Script: runs
- PyTorch Script → Synthetic Data: generates
train_torch.py - Synthetic Data → Input Tensors: converts
train_torch.py - Input Tensors → Linear Score: computes
train_torch.py - Tracked Parameters → Linear Score: weights
train_torch.py - Linear Score → Sigmoid: feeds
train_torch.py - Sigmoid → Binary Cross-Entropy: scores against
train_torch.py - Input Tensors → Binary Cross-Entropy: provides labels
train_torch.py - Binary Cross-Entropy → Autograd Backward: backpropagates
train_torch.py - Autograd Backward → Parameter Update: updates gradients
train_torch.py - Tracked Parameters → Parameter Update: receives updates
train_torch.py - Parameter Update → Training Report: reports parameters
train_torch.py - Binary Cross-Entropy → Training Report: logs loss
train_torch.py