pandas shift converts my column from integer to float.
Question:
shift converts my column from integer to float. It turns out that np.nan is float only. Is there any ways to keep shifted column as integer?
df = pd.DataFrame({"a":range(5)})
df['b'] = df['a'].shift(1)
df['a']
# 0 0
# 1 1
# 2 2
# 3 3
# 4 4
# Name: a, dtype: int64
df['b']
# 0 NaN
# 1 0
# 2 1
# 3 2
# 4 3
# Name: b, dtype: float64
Answers:
Solution for pandas under 0.24:
Problem is you get NaN value what is float, so int is converted to float – see na type promotions.
One possible solution is convert NaN values to some value like 0 and then is possible convert to int:
df = pd.DataFrame({"a":range(5)})
df['b'] = df['a'].shift(1).fillna(0).astype(int)
print (df)
a b
0 0 0
1 1 0
2 2 1
3 3 2
4 4 3
Solution for pandas 0.24+ – check Series.shift:
fill_value object, optional
The scalar value to use for newly introduced missing values. the default depends on the dtype of self. For numeric data, np.nan is used. For datetime, timedelta, or period data, etc. NaT is used. For extension dtypes, self.dtype.na_value is used.
Changed in version 0.24.0.
df['b'] = df['a'].shift(fill_value=0)
You can construct a numpy array by prepending a 0 to all but the last element of column a
df.assign(b=np.append(0, df.a.values[:-1]))
a b
0 0 0
1 1 0
2 2 1
3 3 2
4 4 3
another solution is to use replace() function and type cast
df['b'] = df['a'].shift(1).replace(np.NaN,0).astype(int)
another solution starting from pandas version 0.24.0: simply provide a value for the parameter fill_value:
df['b'] = df['a'].shift(1, fill_value=0)
As of pandas 1.0.0 I believe you have another option, which is to first use convert_dtypes. This converts the dataframe columns to dtypes that support pd.NA, avoiding the issues with NaN.
df = pd.DataFrame({"a":range(5)})
df = df.convert_dtypes()
df['b'] = df['a'].shift(1)
print(df['a'])
# 0 0
# 1 1
# 2 2
# 3 3
# 4 4
# Name: a, dtype: Int64
print(df['b'])
# 0 <NA>
# 1 0
# 2 1
# 3 2
# 4 3
# Name: b, dtype: Int64
I don’t like other answers which may change original dtypes, what if you have float, str in data?
Since we don’t need the first nan row , why not skip it.
I would keep all dtypes and cast back:
dt = df.dtypes
df = df.shift(1).iloc[1:].astype(dt)
shift converts my column from integer to float. It turns out that np.nan is float only. Is there any ways to keep shifted column as integer?
df = pd.DataFrame({"a":range(5)})
df['b'] = df['a'].shift(1)
df['a']
# 0 0
# 1 1
# 2 2
# 3 3
# 4 4
# Name: a, dtype: int64
df['b']
# 0 NaN
# 1 0
# 2 1
# 3 2
# 4 3
# Name: b, dtype: float64
Solution for pandas under 0.24:
Problem is you get NaN value what is float, so int is converted to float – see na type promotions.
One possible solution is convert NaN values to some value like 0 and then is possible convert to int:
df = pd.DataFrame({"a":range(5)})
df['b'] = df['a'].shift(1).fillna(0).astype(int)
print (df)
a b
0 0 0
1 1 0
2 2 1
3 3 2
4 4 3
Solution for pandas 0.24+ – check Series.shift:
fill_value object, optional
The scalar value to use for newly introduced missing values. the default depends on the dtype of self. For numeric data, np.nan is used. For datetime, timedelta, or period data, etc. NaT is used. For extension dtypes, self.dtype.na_value is used.Changed in version 0.24.0.
df['b'] = df['a'].shift(fill_value=0)
You can construct a numpy array by prepending a 0 to all but the last element of column a
df.assign(b=np.append(0, df.a.values[:-1]))
a b
0 0 0
1 1 0
2 2 1
3 3 2
4 4 3
another solution is to use replace() function and type cast
df['b'] = df['a'].shift(1).replace(np.NaN,0).astype(int)
another solution starting from pandas version 0.24.0: simply provide a value for the parameter fill_value:
df['b'] = df['a'].shift(1, fill_value=0)
As of pandas 1.0.0 I believe you have another option, which is to first use convert_dtypes. This converts the dataframe columns to dtypes that support pd.NA, avoiding the issues with NaN.
df = pd.DataFrame({"a":range(5)})
df = df.convert_dtypes()
df['b'] = df['a'].shift(1)
print(df['a'])
# 0 0
# 1 1
# 2 2
# 3 3
# 4 4
# Name: a, dtype: Int64
print(df['b'])
# 0 <NA>
# 1 0
# 2 1
# 3 2
# 4 3
# Name: b, dtype: Int64
I don’t like other answers which may change original dtypes, what if you have float, str in data?
Since we don’t need the first nan row , why not skip it.
I would keep all dtypes and cast back:
dt = df.dtypes
df = df.shift(1).iloc[1:].astype(dt)