Python 如何将包含日期的csv数据加载到numpy数组中?
我尝试不同的东西太久了 如何将包含日期的csv数据加载到numpy数组中? 这是不起作用的。它创建了一条直线,所有应该是直线的东西现在都在一个单元格中Python 如何将包含日期的csv数据加载到numpy数组中?,python,numpy,csv,datetime,Python,Numpy,Csv,Datetime,我尝试不同的东西太久了 如何将包含日期的csv数据加载到numpy数组中? 这是不起作用的。它创建了一条直线,所有应该是直线的东西现在都在一个单元格中 import io import numpy as np import datetime as dt def date_parser(d_bytes): s = d_bytes.decode('utf-8') return np.datetime64(dt.datetime.strptime(s, "%Y-%m-%d %H:%
import io
import numpy as np
import datetime as dt
def date_parser(d_bytes):
s = d_bytes.decode('utf-8')
return np.datetime64(dt.datetime.strptime(s, "%Y-%m-%d %H:%M:%S"))
def read_csv():
five_min_candles_str = """2020-06-01 17:05:00,9506.01,9523.31,9500.0,9514.52
2020-06-01 17:10:00,9513.44,9525.22,9500.32,9522.0
2020-06-01 17:15:00,9521.56,9525.59,9513.75,9523.53
2020-06-01 17:20:00,9523.21,9525.53,9518.78,9524.55
2020-06-01 17:25:00,9524.55,9538.4,9522.93,9528.73
2020-06-01 17:30:00,9528.73,9548.98,9527.95,9543.72
2020-06-01 17:35:00,9542.71,9547.34,9536.57,9543.66
2020-06-01 17:40:00,9543.67,9543.67,9530.0,9531.85
2020-06-01 17:45:00,9530.84,9535.01,9524.1,9526.75
2020-06-01 17:50:00,9526.47,9538.64,9521.87,9534.57
2020-06-01 17:55:00,9534.58,9548.9,9533.04,9546.98
2020-06-01 18:00:00,9548.18,9558.9,9536.99,9556.25
2020-06-01 18:05:00,9556.15,9579.8,9547.7,9574.09
2020-06-01 18:10:00,9575.0,9592.59,9571.3,9573.93
2020-06-01 18:15:00,9573.68,9610.0,9569.6,9597.78
2020-06-01 18:20:00,9597.78,9598.85,9578.0,9591.39
"""
nparray = np.genfromtxt(io.StringIO(five_min_candles_str),
delimiter=',',
dtype=[('Timestamp','datetime64[us]'),
('Open','object'),
('High','object'),
('Low','object'),
('Close','object')],
converters={0: date_parser},
)
print(nparray)
if __name__ == "__main__":
read_csv()
如能提供解决方案或提示,将不胜感激
编辑:
事实证明,它确实已经开始工作了,但在我添加了类型或转换器之后,它变成了元组数组,而我希望它是一个2D数组。原因是一行中的不同类型。看
我将下面的答案标记为正确答案,因为我更喜欢它,因为它不需要对日期进行任何自定义解析,而且与io.StringIO()相比,我更喜欢splitlines()
解决方案
让我们看看numpy
如何处理这些日期字符串。它不像熊猫那样健壮,但是:
In [55]: np.array('2020-06-01 17:05:00', 'datetime64[s]')
Out[55]: array('2020-06-01T17:05:00', dtype='datetime64[s]')
但是看起来不错。日期和时间之间的间隔是正常的(“T”也起作用)
因此,让我们尝试一种全自动的数据类型:
In [56]: data=np.genfromtxt(five_min_candles_str.splitlines(), delimiter=',', dt
...: ype=None, encoding=True)
In [57]: data
Out[57]:
array([('2020-06-01 17:05:00', 9506.01, 9523.31, 9500. , 9514.52),
('2020-06-01 17:10:00', 9513.44, 9525.22, 9500.32, 9522. ),
('2020-06-01 17:15:00', 9521.56, 9525.59, 9513.75, 9523.53),
...
('2020-06-01 18:20:00', 9597.78, 9598.85, 9578. , 9591.39)],
dtype=[('f0', '<U19'), ('f1', '<f8'), ('f2', '<f8'), ('f3', '<f8'), ('f4', '<f8')])
[56]中的:data=np.genfromtxt(five_min_str.splitlines(),delimiter=',,dt
…:ype=None,encoding=True)
在[57]中:数据
出[57]:
阵列([('2020-06-0117:05:00',9506.019523.319500,9514.52),
('2020-06-01 17:10:00', 9513.44, 9525.22, 9500.32, 9522. ),
('2020-06-01 17:15:00', 9521.56, 9525.59, 9513.75, 9523.53),
...
('2020-06-01 18:20:00', 9597.78, 9598.85, 9578. , 9591.39)],
dtype=[('f0','为完整起见,这几乎是一个使用pandas
的单行程序:
from io import StringIO
import pandas as pd
s="""2020-06-01 17:05:00,9506.01,9523.31,9500.0,9514.52
2020-06-01 17:10:00,9513.44,9525.22,9500.32,9522.0
2020-06-01 17:15:00,9521.56,9525.59,9513.75,9523.53
2020-06-01 17:20:00,9523.21,9525.53,9518.78,9524.55
2020-06-01 17:25:00,9524.55,9538.4,9522.93,9528.73
2020-06-01 17:30:00,9528.73,9548.98,9527.95,9543.72
2020-06-01 17:35:00,9542.71,9547.34,9536.57,9543.66
2020-06-01 17:40:00,9543.67,9543.67,9530.0,9531.85
2020-06-01 17:45:00,9530.84,9535.01,9524.1,9526.75
2020-06-01 17:50:00,9526.47,9538.64,9521.87,9534.57
2020-06-01 17:55:00,9534.58,9548.9,9533.04,9546.98
2020-06-01 18:00:00,9548.18,9558.9,9536.99,9556.25
2020-06-01 18:05:00,9556.15,9579.8,9547.7,9574.09
2020-06-01 18:10:00,9575.0,9592.59,9571.3,9573.93
2020-06-01 18:15:00,9573.68,9610.0,9569.6,9597.78
2020-06-01 18:20:00,9597.78,9598.85,9578.0,9591.39"""
df = pd.read_csv(StringIO(s), names=['Timestamp','Open','High','Low','Close'],
parse_dates=[0])
for c in df.columns:
print(f"{c} - {df[c].dtype}")
# Timestamp - datetime64[ns]
# Open - float64
# High - float64
# Low - float64
# Close - float64
你为什么不使用pandas
?因为我想迭代我学到的行,这是pandas
的反模式。如果你能找到矢量化函数,行迭代是不可取的,但它随时都可以供你使用。在这种情况下,pandas
可以帮助你简化csv文件的获取。我使用的数据如果我能在这里找到一个解决方案,我想numpy仍然是首选的方式。但我现在对这两种方式都没有太多的经验来真正判断。在pandas
中的行迭代并不比numpy
中的行迭代差。我们在pandas
中看到了很多这样的情况ode>apply
。为了提高速度,在整个数据帧/数组上进行操作是很好的,但是有些任务可能太复杂了。是的,这对我来说也非常简单。parse_dates=True
就够了。好的,明白了。IDE中的数据看起来有误,但实际上还可以。我仍然更喜欢你的答案不需要解析。
In [58]: dt = [('f0', 'datetime64[s]'), ('f1', '<f8'), ('f2', '<f8'), ('f3', '<f8'), ('f4', '<f8')]
In [59]: data=np.genfromtxt(five_min_candles_str.splitlines(), delimiter=',', dtype=dt, encoding=True)
In [60]: data
Out[60]:
array([('2020-06-01T17:05:00', 9506.01, 9523.31, 9500. , 9514.52),
('2020-06-01T17:10:00', 9513.44, 9525.22, 9500.32, 9522. ),
('2020-06-01T17:15:00', 9521.56, 9525.59, 9513.75, 9523.53),
('2020-06-01T17:20:00', 9523.21, 9525.53, 9518.78, 9524.55),
...
('2020-06-01T18:20:00', 9597.78, 9598.85, 9578. , 9591.39)],
dtype=[('f0', '<M8[s]'), ('f1', '<f8'), ('f2', '<f8'), ('f3', '<f8'), ('f4', '<f8')])
from io import StringIO
import pandas as pd
s="""2020-06-01 17:05:00,9506.01,9523.31,9500.0,9514.52
2020-06-01 17:10:00,9513.44,9525.22,9500.32,9522.0
2020-06-01 17:15:00,9521.56,9525.59,9513.75,9523.53
2020-06-01 17:20:00,9523.21,9525.53,9518.78,9524.55
2020-06-01 17:25:00,9524.55,9538.4,9522.93,9528.73
2020-06-01 17:30:00,9528.73,9548.98,9527.95,9543.72
2020-06-01 17:35:00,9542.71,9547.34,9536.57,9543.66
2020-06-01 17:40:00,9543.67,9543.67,9530.0,9531.85
2020-06-01 17:45:00,9530.84,9535.01,9524.1,9526.75
2020-06-01 17:50:00,9526.47,9538.64,9521.87,9534.57
2020-06-01 17:55:00,9534.58,9548.9,9533.04,9546.98
2020-06-01 18:00:00,9548.18,9558.9,9536.99,9556.25
2020-06-01 18:05:00,9556.15,9579.8,9547.7,9574.09
2020-06-01 18:10:00,9575.0,9592.59,9571.3,9573.93
2020-06-01 18:15:00,9573.68,9610.0,9569.6,9597.78
2020-06-01 18:20:00,9597.78,9598.85,9578.0,9591.39"""
df = pd.read_csv(StringIO(s), names=['Timestamp','Open','High','Low','Close'],
parse_dates=[0])
for c in df.columns:
print(f"{c} - {df[c].dtype}")
# Timestamp - datetime64[ns]
# Open - float64
# High - float64
# Low - float64
# Close - float64