NumPy是否有类似Pandas的select_dtypes方法？实现按类型筛选

NumPy doesn’t have a built-in select_dtypes function like Pandas, but we can easily replicate this functionality using structured arrays—NumPy’s way of handling tabular data with mixed data types. Here’s a step-by-step implementation:

Step 1: Convert your Pandas DataFrame to a NumPy structured array #

Standard NumPy arrays are homogeneous (all elements share the same dtype), so we first convert the DataFrame to a structured array (which supports mixed dtypes) using df.to_records():

import pandas as pd
import numpy as np

# Create your sample DataFrame
df = pd.DataFrame({'a': [1, 2] * 3, 'b': [True, False] * 3, 'c': [1.0, 2.0] * 3})

# Convert to structured array (exclude the index for cleaner output)
np_structured_arr = df.to_records(index=False)

Step 2: Build the dtype-filtering function #

We’ll create a custom function that filters fields in the structured array based on their dtypes, mirroring Pandas’ select_dtypes behavior:

def select_dtypes_numpy(arr, include=None, exclude=None):
    # Set default empty lists if parameters aren't provided
    include = include or []
    exclude = exclude or []
    
    # Helper to convert dtype strings (like 'float64') to actual NumPy dtype objects
    def resolve_dtype(dt):
        return np.dtype(dt) if isinstance(dt, str) else dt
    
    # Normalize include/exclude lists to use consistent dtype objects
    include_dtypes = [resolve_dtype(dt) for dt in include]
    exclude_dtypes = [resolve_dtype(dt) for dt in exclude]
    
    # Get all field names and their corresponding dtypes
    field_details = [(name, arr.dtype[name]) for name in arr.dtype.names]
    
    # Filter fields that match our rules
    selected_fields = []
    for field_name, field_dtype in field_details:
        # Skip if the field is explicitly excluded
        if exclude_dtypes and field_dtype in exclude_dtypes:
            continue
        # Keep the field if it's included (or if include list is empty)
        if not include_dtypes or field_dtype in include_dtypes:
            selected_fields.append(field_name)
    
    # Return the filtered structured array
    return arr[selected_fields]

Step 3: Use the function to replicate your Pandas example #

Now we can use this function to select only float64 fields, just like your Pandas code:

# Select float64 fields
float_only_arr = select_dtypes_numpy(np_structured_arr, include=['float64'])
print(float_only_arr)

Output: #

[(1.,) (2.,) (1.,) (2.,) (1.,) (2.,)]

If you want to convert this back to a Pandas DataFrame for readability, simply run:

pd.DataFrame(float_only_arr)

This will give you the exact same output as df.select_dtypes(include=['float64']).

Extra features #

• Use the exclude parameter to omit specific dtypes, e.g., select_dtypes_numpy(arr, exclude=['bool']) would remove the boolean field from your sample data.
• The function works with both dtype strings (like 'int64') and NumPy dtype objects (like np.int64).

内容的提问来源于stack exchange，提问作者Student