feat: Implement interactive 3D feature visualization

[bshihab]

New Feature: Interactive 3D Feature View

This pull request introduces a new 3D feature view, allowing for interactive visualization of spike features in a 3D scatter plot. This provides a more intuitive way to explore the relationships between different principal components and other spike attributes.

The rotation is powered by a standard Yaw-Pitch camera model. Holding Shift while dragging the mouse controls the camera's pitch (up/down rotation around the X-axis) and yaw (left/right rotation around the Y-axis) around the data's center.

Technically, this is implemented in the _create_rotation_matrix function, which builds two separate 3x3 rotation matrices, R_x (for pitch) and R_y (for yaw). These are combined via matrix multiplication (R_x @ R_y) into a single transformation matrix. Applying yaw first and then pitch results in a smooth and intuitive orbiting behavior. This final matrix transforms the 3D feature coordinates before they are projected onto the 2D screen, creating a responsive exploration experience.

Key Features

• Interactive 3D Scatter Plot: A new Feature3DView has been added to display features in a 3D space.

• Rotation and Zoom: Users can rotate, pan, and zoom the 3D view to inspect the data from any angle.

• Customizable Axes: The X, Y, and Z axes can be mapped to different principal components or other spike attributes.

• Lasso Selection: Spikes can be selected in the 3D view using a lasso tool for further analysis or clustering.

Implementation Details

• Created a new Feature3DView class in phy/cluster/views/featureview3d.py.

• Integrated the new view into the TemplateGUI in phy/apps/template/gui.py.

• Updated phy/gui/qt.py and other files to support the new view.

How to Test This Branch

These instructions outline how to set up a clean conda environment to test the changes in this PR. This assumes a base installation of Miniconda and an environment created with Python 3.11.

1. Install Git in Conda
If git is not already available in the environment, install it:
conda install git -y

2. Clone This Specific Branch
This command clones only the feature-3d-view branch into a new folder named phy-feature-3d-view.
(Note: Replace C:\Users\YourUsername\Desktop with your desired location.)
First, navigate to your desired directory:
cd C:\Users\YourUsername\Desktop
Then, run the clone command:
git clone --branch feature-3d-view --single-branch https://github.com/bshihab/phy.git phy-feature-3d-view

3. Install Dependencies
Navigate into the new directory and install the package in editable mode.
A critical step is to pin joblib to version 1.2.0. I found that recent versions of joblib are incompatible and cause the GUI to crash on launch due to a removed keyword argument in the Memory class.
First, change into the repository folder:
cd phy-feature-3d-view
Ensure pip and its core tools are up to date:
python -m pip install --upgrade pip setuptools wheel
Install the package in editable mode. This will install most dependencies from requirements.txt:
python -m pip install -e .
Install the full set of core dependencies with pinned versions to ensure stability:
python -m pip install joblib==1.2.0 pyopengl==3.1.6 numpy traitlets matplotlib click PyQt5 PyQtWebEngine qtconsole scipy toolz requests dask h5py cython
After these steps, the environment is ready for testing the new features.

[bshihab]

Hi @rossant , I would be grateful if you could review this pull request for a new 3D feature view. I've detailed the implementation and testing steps in the description above. Looking forward to your feedback!

[rossant]

Sorry for the delay, and thank you for your contribution!
I’m not able to test or review it right now, maybe @zm711 could take a look?
I’d be happy to grant merge permissions to anyone who has more bandwidth than I do to help maintain this repo.
Thanks again.