HDF5 min_itemsize 错误：ValueError: Trying to store a string with len [##] in [y] column but this column has a limit of [##]！

Question

使用pandas.HDFStore().append()后出现以下错误

ValueError: Trying to store a string with len [150] in [values_block_0] column but  this column has a limit of [127]!

Consider using min_itemsize to preset the sizes on these columns

我正在创建一个 pandas DataFrame 并将其附加到 HDF5 文件中，如下所示：

import pandas as pd

store = pd.HDFStore("test1.h5", mode='w')

hdf_key = "one_key"

columns = ["col1", "col2", ... ]

df = pd.Dataframe(...)
df.col1 = df.col1.astype(str)
df.col2 = df.col2astype(int)
df.col3 = df.col3astype(str)
.... 
store.append(hdf_key, df, data_column=columns, index=False)

我收到上述错误：“ValueError: Trying to store a string with len [150] in [values_block_0] column but this column has a limit of [127]！”

之后，我执行代码：

store.get_storer(hdf_key).table.description

哪个输出

{
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "values_block_0": StringCol(itemsize=127, shape=(5,), dflt=b'', pos=1),
  "values_block_1": Int64Col(shape=(5,), dflt=0, pos=2),
  "col1": StringCol(itemsize=20, shape=(), dflt=b'', pos=3),
  "col2": StringCol(itemsize=39, shape=(), dflt=b'', pos=4)}

values_block_0 和 values_block_1 是什么？

所以，在这个 StackOverflow Pandas pytable: how to specify min_itemsize of the elements of a MultiIndex 之后，我尝试了

store.append(hdf_key, df, data_column=columns, index=False,  min_itemsize={"values_block_0":250})

这不起作用——现在我得到这个错误：

ValueError: Trying to store a string with len [250] in [values_block_0] column but  this column has a limit of [127]!

Consider using min_itemsize to preset the sizes on these columns

我做错了什么？

编辑：此代码从filename.py 产生错误ValueError: min_itemsize has the key [values_block_0] which is not an axis or data_column

import pandas as pd
store = pd.HDFStore("test1.h5", mode='w')
hdf_key = "one_key"

my_columns = ["col1", "col2", ... ]

df = pd.Dataframe(...)
df.col1 = df.col1.astype(str)
df.col2 = df.col2astype(int)
df.col3 = df.col3astype(str)
.... 
store.append(hdf_key, df, data_column=my_columns, index=False, min_itemsize={"values_block_0":350})

这是完整的错误：

(python-3) -bash:1008 $ python filename.py
Traceback (most recent call last):
  File "filename.py", line 50, in <module>
    store.append(hdf_key, dicts_into_df,  data_column=my_columns, index=False, min_itemsize={'values_block_0':350})
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 970, in append
    **kwargs)
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 1315, in _write_to_group
    s.write(obj=value, append=append, complib=complib, **kwargs)
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 4263, in write
    obj=obj, data_columns=data_columns, **kwargs)
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 3853, in write
    **kwargs)
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 3535, in create_axes
    self.validate_min_itemsize(min_itemsize)
  File "/path/lib/python-3/lib/python3.5/site-packages/pandas/io/pytables.py", line 3174, in validate_min_itemsize
    "data_column" % k)
ValueError: min_itemsize has the key [values_block_0] which is not an axis or data_column

Answer 1

更新：

您拼错了data_columns 参数：data_column - 它应该是data_columns。结果，您的 HDF 存储中没有任何索引列，并且添加了 HDF 存储values_block_X：

In [70]: store = pd.HDFStore(r'D:\temp\.data\my_test.h5')

拼写错误的参数将被忽略：

In [71]: store.append('no_idx_wrong_dc', df, data_column=df.columns, index=False)

In [72]: store.get_storer('no_idx_wrong_dc').table
Out[72]:
/no_idx_wrong_dc/table (Table(10,)) ''
  description := {
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "values_block_0": Float64Col(shape=(1,), dflt=0.0, pos=1),
  "values_block_1": Int64Col(shape=(1,), dflt=0, pos=2),
  "values_block_2": StringCol(itemsize=30, shape=(1,), dflt=b'', pos=3)}
  byteorder := 'little'
  chunkshape := (1213,)

与以下相同：

In [73]: store.append('no_idx_no_dc', df, index=False)

In [74]: store.get_storer('no_idx_no_dc').table
Out[74]:
/no_idx_no_dc/table (Table(10,)) ''
  description := {
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "values_block_0": Float64Col(shape=(1,), dflt=0.0, pos=1),
  "values_block_1": Int64Col(shape=(1,), dflt=0, pos=2),
  "values_block_2": StringCol(itemsize=30, shape=(1,), dflt=b'', pos=3)}
  byteorder := 'little'
  chunkshape := (1213,)

让我们正确拼写：

In [75]: store.append('no_idx_dc', df, data_columns=df.columns, index=False)

In [76]: store.get_storer('no_idx_dc').table
Out[76]:
/no_idx_dc/table (Table(10,)) ''
  description := {
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "value": Float64Col(shape=(), dflt=0.0, pos=1),
  "count": Int64Col(shape=(), dflt=0, pos=2),
  "s": StringCol(itemsize=30, shape=(), dflt=b'', pos=3)}
  byteorder := 'little'
  chunkshape := (1213,)

旧答案：

AFAIK 你可以有效地设置min_itemsize参数在第一个只追加。

演示：

In [33]: df
Out[33]:
   num                 s
0   11  aaaaaaaaaaaaaaaa
1   12    bbbbbbbbbbbbbb
2   13     ccccccccccccc
3   14       ddddddddddd

In [34]: store = pd.HDFStore(r'D:\temp\.data\my_test.h5')

In [35]: store.append('test_1', df, data_columns=True)

In [36]: store.get_storer('test_1').table.description
Out[36]:
{
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "num": Int64Col(shape=(), dflt=0, pos=1),
  "s": StringCol(itemsize=16, shape=(), dflt=b'', pos=2)}

In [37]: df.loc[4] = [15, 'X'*200]

In [38]: df
Out[38]:
   num                                                  s
0   11                                   aaaaaaaaaaaaaaaa
1   12                                     bbbbbbbbbbbbbb
2   13                                      ccccccccccccc
3   14                                        ddddddddddd
4   15  XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX...

In [39]: store.append('test_1', df, data_columns=True)
...
skipped
...
ValueError: Trying to store a string with len [200] in [s] column but
this column has a limit of [16]!
Consider using min_itemsize to preset the sizes on these columns    

现在使用min_itemsize，但仍附加到现有的store 对象：

In [40]: store.append('test_1', df, data_columns=True, min_itemsize={'s':250})
...
skipped
...
ValueError: Trying to store a string with len [250] in [s] column but
this column has a limit of [16]!
Consider using min_itemsize to preset the sizes on these columns

如果我们要在 store 中创建一个新对象，则以下工作：

In [41]: store.append('test_2', df, data_columns=True, min_itemsize={'s':250})

检查列大小：

In [42]: store.get_storer('test_2').table.description
Out[42]:
{
  "index": Int64Col(shape=(), dflt=0, pos=0),
  "num": Int64Col(shape=(), dflt=0, pos=1),
  "s": StringCol(itemsize=250, shape=(), dflt=b'', pos=2)}

Answer 2

我大约在将 Pandas 从 18.1 更新到 22.0 的同时开始收到此错误（尽管这可能无关）。

我通过手动读取数据帧来修复现有 HDF5 文件中的错误，然后为错误中提到的列写入一个具有更大 min_itemsize 的新 HDF5 文件：

filename_hdf5 = "C:\test.h5"
df = pd.read_hdf(filename_hdf5, 'table_name')
hdf = HDFStore(filename_hdf5)
hdf.put('table_name', df, format='table', data_columns=True, min_itemsize={'ColumnNameMentionedInError': 10})
hdf.close()

然后我更新了现有代码以在创建密钥时设置min_itemsize。

---

专家补充

发生错误是因为尝试将更多行附加到现有数据帧，其固定列宽对于新数据来说太窄。固定列宽最初是根据第一次写入数据帧时列中最长的字符串设置的。

我认为 pandas 应该透明地处理这个错误，而不是为所有未来的附加操作留下一个有效的定时炸弹。这个问题可能需要数周甚至数年才能浮出水面。