我想在任意一个Dask系列上调用.value_counts()
,并且如果该系列包含不可散列的类型,我想将该系列转换为类型 string 。如果不需要,我不想将系列转换为字符串。我也不想在致电.compute()
之前先致电.value_counts()
。我尝试过
df = pd.DataFrame({"a":[[1], ["foo"], ["foo", "bar"]]})
df = dd.from_pandas(df, npartitions=1)
srs = df["a"]
try:
val_counts = srs.value_counts()
except TypeError:
srs = srs.astype(str)
val_counts = srs.value_counts()
val_counts.compute()
出现错误
TypeError:不可散列的类型:“列表”
和
df = pd.DataFrame({"a":[[1], ["foo"], ["foo", "bar"]]})
df = dd.from_pandas(df, npartitions=1)
srs = df["a"]
def func(srs):
try:
val_counts = srs.value_counts()
except TypeError:
srs = srs.astype(str)
val_counts = srs.value_counts()
return val_counts
val_counts = dask.compute(func(srs))
给出相同的错误。
我也尝试过
df = pd.DataFrame({"a":[[1], ["foo"], ["foo", "bar"]]})
df = dd.from_pandas(df, npartitions=1)
srs = df["a"]
if srs.apply(lambda y: isinstance(y, list), meta=srs).any():
srs = srs.astype(str)
srs.value_counts().compute()
出现错误
TypeError:尝试将dd.Scalar
转换为布尔值。
答案 0 :(得分:2)
也许首先将列表转换成可哈希的东西(如元组)?
s.apply(tuple).value_counts() ?