gradfield 0.4.0

Gradient field & loss landscape visualizer for PyTorch and NumPy

gradfield 🔥

Gradient field & loss landscape visualizer for PyTorch and NumPy

Visualize any function's loss surface, gradient arrows, and optimization trajectories — in an interactive 3D HTML chart. No server, no config, one function call.

---

Install

pip install gradfield

With PyTorch support:

pip install gradfield torch

---

Quick Start

Any Python function

from gradfield import sample_grid, render

field = sample_grid(
    lambda x, y: (1 - x)**2 + 100 * (y - x**2)**2,  # Rosenbrock
    x_range=(-2, 2),
    y_range=(-1, 3),
    resolution=80,
)

render(field, title="Rosenbrock Banana")
# Opens an interactive HTML in your browser ✓

PyTorch model loss landscape

import torch.nn as nn
from gradfield import model_loss_fn, sample_grid, render

landscape = model_loss_fn(
    model=my_model,
    loss_fn=nn.CrossEntropyLoss(),
    inputs=X_batch,
    targets=y_batch,
    scale=0.5,
)

field = sample_grid(landscape, resolution=50)
render(field, title="My Model Loss Landscape")

With optimization trajectories

import numpy as np

traj = np.array([[2.0, 2.0], [1.5, 1.5], [0.5, 0.5], [0.0, 0.0]])

render(field, trajectories=[traj], title="Gradient Descent Path")

---

API Reference

sample_grid(fn, x_range, y_range, resolution)

Sample a function over a 2D grid using NumPy central differences for gradients.

sample_grid_torch(fn, x_range, y_range, resolution)

Same as above but uses torch.autograd for exact gradients.

model_loss_fn(model, loss_fn, inputs, targets, scale)

Wraps a PyTorch model into a 2D landscape function using random filter-normalized directions (Li et al. 2018).

render(field, trajectories, title, output, colorscale, show_gradients)

Render the sampled field as an interactive Plotly HTML with:

• 3D surface plot
• 2D contour map
• Gradient vector field overlay
• Optimization trajectory paths

---

Examples

cd examples
python example_numpy.py       # Rosenbrock + gradient descent
python example_torch_model.py # MLP on XOR loss landscape

---

Roadmap

• C++ accelerated grid sampler via pybind11
• 1D loss curve along a direction
• SGD / Adam / RMSProp built-in trajectory recorders
• Jupyter widget support
• Multi-trajectory comparison

---

License

MIT — built with ❤️ by Aman