Polars列使用unnest()触发SchemaError问题求助

我是Polars新手，尝试对DataFrame中的列使用.unnest()方法。我生成了名为"processed"的字符串列表列，列表包含序列和百分比字符串，希望将其拆分为序列列和对应覆盖率列。此前相同场景下unnest()可正常运行，但此次报错，怀疑是"processed"列的生成存在问题，却无法定位原因。

代码 #

import polars as pl
from functools import reduce

# 替换蛋白序列中的肽段为短横线
def replace_peptide(protein_sequence, peptide):
    return protein_sequence.replace(peptide, '-' * len(peptide))

# 合并多个替换后的序列，标记所有肽段覆盖位置
def combine_sequences(seq1, seq2):
    return ''.join(['-' if a == '-' or b == '-' else a for a, b in zip(seq1, seq2)])

# 处理单行数据，生成标记后的序列和覆盖率
def process_row(row):
    protein_sequence = row['sequence']
    protein_sequence_len = len(protein_sequence)
    peptides = row['peptide']
    modified_sequences = [replace_peptide(protein_sequence, peptide) for peptide in peptides]
    final_sequence = reduce(combine_sequences, modified_sequences)
    coverage = str('{:.2f}%').format(final_sequence.count('-')/protein_sequence_len*100)
    return final_sequence, coverage
# -----------------------------------------------------------------------------------------------# 
# 生成初始DataFrame
coverage_fasta=pl.DataFrame({"protein_id": ["A0A024RBG1", "A0A087X1C5", "A0A0B4J2F0"],
                             "sequence": ["MMKFKPNQTRTYDREGFKKRAACLCFRSEQEDEVLLVSSSRYPDQWIVPG",
                                          "MGLEALVPLAMIVAIFLLLVDLMHRHQRWAARYPPGPLPLPGLGNLLHVD",
                                            "MFRRLTFAQLLFATVLGIAGGVYIFQPVFEQYAKDQKELKEKMQLVQESE"],
                             "peptide": [["MMKFKPNQT", "FKKRAA", "SSSRYPDQ"],
                                         ["EALVPLAM", "AQLLFATVLGIAG", "QPVFEQYAKDQ"], 
                                         ["RRLTFAQLL", "LTFAQLLFATVLGIAGG", "QYAKDQKEL"]
                            ]}
)

# 生成包含标记序列和覆盖率的列表列
coverage_fasta = coverage_fasta.with_columns(
    pl.struct(["sequence", "peptide"]).map_elements(lambda row: process_row(row),
                                                    return_dtype=pl.List(pl.String())
                                                    ).alias("processed"))
display(coverage_fasta)

# 尝试拆分processed列，报错位置
coverage_fasta = coverage_fasta.with_columns(pl.col("processed")).unnest("processed")
display(coverage_fasta)

报错信息 #

执行unnest("processed")时触发类型错误，提示无法对List(String)类型的列执行unnest操作。

错误根源 #

processed列是**List(String)**类型（每行是包含2个字符串的列表），而Polars的unnest()方法仅支持展开嵌套的Struct列，或者将List[Struct]类型的列拆分为多列。直接对普通List列使用unnest()会因类型不匹配报错。

快速修复方案 #

以下两种方法都能实现将列表拆分为两列的需求：

方案1：转Struct后再展开

先把List列转为Struct类型（自定义列名），再用unnest()展开：

# 替换原报错的unnest代码
coverage_fasta = coverage_fasta.with_columns(
    pl.col("processed").list.to_struct(fields=["dashed_sequence", "coverage"])
).unnest("processed")
display(coverage_fasta)

方案2：直接提取列表元素

通过list.get(index)直接提取列表中的第0、1位元素生成新列：

# 替换原报错的unnest代码
coverage_fasta = coverage_fasta.with_columns(
    pl.col("processed").list.get(0).alias("dashed_sequence"),
    pl.col("processed").list.get(1).alias("coverage")
).drop("processed")
display(coverage_fasta)

进阶优化（更规范的写法） #

如果在生成processed列时直接返回Struct类型，后续可以直接unnest()，步骤更简洁：

1. 修改process_row函数返回字典：

def process_row(row):
    protein_sequence = row['sequence']
    protein_sequence_len = len(protein_sequence)
    peptides = row['peptide']
    modified_sequences = [replace_peptide(protein_sequence, peptide) for peptide in peptides]
    final_sequence = reduce(combine_sequences, modified_sequences)
    coverage = str('{:.2f}%').format(final_sequence.count('-')/protein_sequence_len*100)
    # 返回字典而非元组
    return {"dashed_sequence": final_sequence, "coverage": coverage}

2. 生成processed列时指定返回类型为Struct：

coverage_fasta = coverage_fasta.with_columns(
    pl.struct(["sequence", "peptide"]).map_elements(
        lambda row: process_row(row),
        return_dtype=pl.Struct([
            pl.Field("dashed_sequence", pl.String),
            pl.Field("coverage", pl.String)
        ])
    ).alias("processed")
)

3. 直接执行unnest()：

coverage_fasta = coverage_fasta.unnest("processed")
display(coverage_fasta)

内容的提问来源于stack exchange，提问作者Giamma Fer