Visualizing relationships between feature spaces

Visualizing relationships between feature spaces#

When analyzing feature measurements and dimensionality reduction results, e.g. involving a UMAP, one often switches between those spaces. In this notebook we demonstrate how one can interpolated between these spaces.

import pandas as pd
import seaborn as sns

First, we load a pandas dataframe of a .csv file containing some measurements, two UMAP columns and a manual selection of some data points.

df = pd.read_csv('../../data/blobs_statistics_with_umap.csv')

df.head()
Unnamed: 0 area mean_intensity minor_axis_length major_axis_length eccentricity extent feret_diameter_max equivalent_diameter_area bbox-0 bbox-1 bbox-2 bbox-3 UMAP1 UMAP2 selection
0 0 422 192.379147 16.488550 34.566789 0.878900 0.586111 35.227830 23.179885 0 11 30 35 7.685868 3.738014 False
1 1 182 180.131868 11.736074 20.802697 0.825665 0.787879 21.377558 15.222667 0 53 11 74 8.768059 7.172123 True
2 2 661 205.216339 28.409502 30.208433 0.339934 0.874339 32.756679 29.010538 0 95 28 122 6.129949 3.477662 False
3 3 437 216.585812 23.143996 24.606130 0.339576 0.826087 26.925824 23.588253 0 144 23 167 5.410176 4.095456 False
4 4 476 212.302521 19.852882 31.075106 0.769317 0.863884 31.384710 24.618327 0 237 29 256 5.280250 3.861278 False

We then visualize the two UMAP columns and two other columns of features of interest.

sns.scatterplot(data=df, x='UMAP1', y='UMAP2', hue='selection')

<Axes: xlabel='UMAP1', ylabel='UMAP2'>
../_images/2136692726642ba823a9976c27237673a550764531c17dba2848bd5f15547acd.png 
sns.scatterplot(data=df, x='area', y='feret_diameter_max', hue='selection')

<Axes: xlabel='area', ylabel='feret_diameter_max'>
../_images/c5541190a5cf8022a0a081cf8b6e3d6e745edff0413d95ae65f97565ac98c180.png

Interpolating between feature spaces#

Drawing an animation of plots that are interpolated between the two plots above can be done using Python. As the procedure is a bit complicated, we asked a language model to write this code for us. Manual adaption of the code was done to make the video play back-and-forth and to modify frame delay. Hence, only the last code line was written by a human.

from bia_bob import bob
bob.initialize(model="gpt-4o-2024-08-06", vision_model="gpt-4o-2024-08-06")
This notebook may contain text, code and images generated by artificial intelligence. Used model: gpt-4o-2024-08-06, vision model: gpt-4o-2024-08-06, endpoint: None, bia-bob version: 0.22.0.. Do not enter sensitive or private information and verify generated contents according to good scientific practice. Read more: https://github.com/haesleinhuepf/bia-bob#disclaimer 
import numpy as np
from sklearn.preprocessing import MinMaxScaler
from io import BytesIO
from skimage.io import imread
import matplotlib.pyplot as plt
import stackview

# Normalize dataframe columns to range between 0 and 1
scaler = MinMaxScaler()
df_normalized = df.copy()
df_normalized[df_normalized.columns] = scaler.fit_transform(df_normalized[df_normalized.columns])

# Create initial plots
fig, ax = plt.subplots(figsize=(8, 6))

# Storage for images
plot_images = []

# Interpolate and generate plots
for i in np.linspace(0, 1, 12): # 12 steps including 0 and 1
    ax.clear()
    x_col = (1 - i) * df_normalized['area'] + i * df_normalized['UMAP1']
    y_col = (1 - i) * df_normalized['feret_diameter_max'] + i * df_normalized['UMAP2']
    
    sns.scatterplot(x=x_col, y=y_col, hue=df['selection'], ax=ax)
    ax.set_title(f"Interpolation {i:.1f}")
    
    buf = BytesIO()
    plt.savefig(buf, format='png')
    buf.seek(0)
    plot_image = imread(buf)
    plot_images.append(plot_image)

plt.close(fig) 

# Use Stackview to animate the sequence of plots
stackview.animate(np.stack(plot_images + plot_images[::-1]))

c:\structure\code\stackview\stackview\_animate.py:64: UserWarning: The image is quite large (> 10 MByte) and might not be properly shown in the notebook when rendered over the internet. Consider subsampling or cropping the image for visualization purposes.
  warnings.warn("The image is quite large (> 10 MByte) and might not be properly shown in the notebook when rendered over the internet. Consider subsampling or cropping the image for visualization purposes.")
Contents