Slider in `scatter_3d` and `scatter` makes some data points go missing

运行 plot_scatter_3d_mwe 中提供的 Python MWE，并比较动画 2D 和 3D 散点图在各个阈值帧之间的行为。跟踪 animation_frame 和 animation_group 在生成的图形中的表示方式；当四个类别及其数据点在滑块的每个位置都仍然可用时，即视为完成。

Only 2 out of 4 categories are plotted when I use a slider. Other data points do not appear at all. When I slide the slider, different categories are plotted. E.g. in the MWE below, only TP and FP show up when the slider is below 0.9. At 0.9 only TN and FN show up.

This behavior also happens for both 2d and 3d scatter plots. See the MWE below for 3d.

import numpy as np
import pandas as pd
import plotly.express as px

def plot_scatter_3d_mwe():
    # Create a small DataFrame with fake data
    data = {
        'Dim1': np.random.rand(10),
        'Dim2': np.random.rand(10),
        'Dim3': np.random.rand(10),
        'due': [1, 0, 1, 0, 1, 0, 1, 0, 1, 0],
        'serial_number': range(10),
        'predicted_probabilities': [0.9, 0.8, 0.4, 0.2, 0.6, 0.7, 0.1, 0.5, 0.3, 0.95]
    }

    df = pd.DataFrame(data)
    thresholds = np.arange(0, 1.1, 0.1)
    all_frames = []

    for threshold in thresholds:
        # Recalculate predictions based on the threshold
        predicted = (df['predicted_probabilities'] >= threshold).astype(int)
        
        # Create the 4 categories for coloring: TP, TN, FP, FN
        conditions = [
            (df['due'] == 1) & (predicted == 1),  # TP
            (df['due'] == 0) & (predicted == 0),  # TN
            (df['due'] == 0) & (predicted == 1),  # FP
            (df['due'] == 1) & (predicted == 0),  # FN
        ]
        categories = ['TP', 'TN', 'FP', 'FN']
        
        # Assign the categories to a new column
        df['category'] = np.select(conditions, categories, default='Unknown')
        df['threshold'] = threshold  # Add threshold as a column for animation frame
        
        all_frames.append(df.copy())

    # Concatenate all frames for animation
    df_all_frames = pd.concat(all_frames)

    # Plot the scatter 3D with the categories as color and animate over thresholds
    fig = px.scatter_3d(df_all_frames,
                        x='Dim1', y='Dim2', z='Dim3',
                        color='category',
                        animation_frame='threshold',
                        animation_group='serial_number')

    fig.show()

# Call the function
plot_scatter_3d_mwe()