如何在Python的tsfresh中仅提取方差与标准差特征？

Hey there! I get that when you're starting out with tsfresh, narrowing down to exactly the features you need can feel a bit tricky. Let me break down how to adjust your code to only compute F_x__variance and F_x__standard_deviation:

The Core Idea #

Instead of using ComprehensiveFCParameters() (which enables all possible features), you'll define a custom settings dictionary that explicitly specifies only the features you want for your F_x column. TSFresh uses this dictionary to know exactly which calculations to run.

Modified Code #

Here's the adjusted version of your code:

from tsfresh import extract_relevant_features

# Define your custom settings: only keep variance and standard_deviation for F_x
custom_settings = {
    'F_x': {
        'variance': None,       # Use default parameters for variance
        'standard_deviation': None  # Use default parameters for standard deviation
    }
}

# Now extract only the specified features
features_filtered_direct = extract_relevant_features(
    df, 
    y, 
    column_id='id', 
    column_sort='time',
    default_fc_parameters=custom_settings  # Pass your custom settings here
)

What's Happening Here? #

• The custom_settings dictionary tells tsfresh: "For the column F_x, only calculate the variance and standard_deviation features."
• Setting the feature values to None means we use the default parameters for those features (neither of these require extra configuration, so this works perfectly).
• If you ever wanted to add parameters to a feature (like adjusting bins for a histogram feature), you'd replace None with a dictionary of parameters instead.

Bonus: If You Need Other Columns' Features Too #

If you have other time series columns and want to keep their default features but only limit F_x to these two, you can start with a base settings object and modify it:

from tsfresh import extract_relevant_features
from tsfresh.feature_extraction import ComprehensiveFCParameters

# Start with all default features
settings = ComprehensiveFCParameters()
# Override F_x to only include your desired features
settings['F_x'] = {'variance': None, 'standard_deviation': None}

# Then use this modified settings object in extract_relevant_features
features_filtered_direct = extract_relevant_features(
    df, 
    y, 
    column_id='id', 
    column_sort='time',
    default_fc_parameters=settings
)

That should give you exactly the two features you're looking for without the extra 95! Let me know if you run into any snags.

内容的提问来源于stack exchange，提问作者Humaun Rashid Nayan