I'm trying to get the datetime type to be used in the numpy array (datep) here. I have tried to approaches for the function bytespdates2num.
First is:
def bytespdates2num(fmt,encoding = 'utf-8'):
def bytesconverter(b):
s = b.decode(encoding)
return mdate.datestr2num(s)
return bytesconverter
Second is:
def bytespdates2num(fmt, encoding = 'utf-8'):
def bytesconverter(b):
s = b.decode(encoding)
return datetime.datetime.strptime(s)
return bytesconverter
My code is:
import urllib
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
import numpy as np
import ssl
import requests
import json
import datetime
#First or second approach
def bytespdates2num(fmt, encoding = 'utf-8'):
def bytesconverter(b):
#makes it into string from utf-8 encoding format
s = b.decode(encoding)
return datetime.datetime.strptime(s,fmt)
return bytesconverter
url1 = 'https://pythonprogramming.net/yahoo_finance_replacement'
data = urllib.request.urlopen(url1,context = None).read().decode()
stockprices = list()
stocksplitdata = data.split('\n')
for line in stocksplitdata:
stockprices.append(line)
date,openp,highp,lowp,closep,adjust,vol = np.loadtxt(stockprices[1:],delimiter = ',',\
unpack = True,\
converters = {0:bytespdates2num('%Y-%m-%d')})
While the first approach works and I can proceed to plot the matplotlib graph using datep as the xaxis, the second approach fails giving float() argument must be a string or a number, not 'datetime.datetime . However while debugging, running the datetime.datetime.strptime(s) line on the command line gives the datetime object of s . Why is this occurring? The datetime approach too turns a string format date into a date time format and seems more straight forward too.
Even though you specify a converter, you still need to specify a dtype.
I tried to recreate your case with a simple input (WHY DIDN'T YOU DO THIS FOR US???)
In [20]: txt = """2011-01-23
...: 2020-03-23"""
Your second converter (one of your's is missing the fmt):
In [21]: def bytespdates2num(fmt, encoding = 'utf-8'):
...: def bytesconverter(b):
...: s = b.decode(encoding)
...: return datetime.datetime.strptime(s, fmt)
...: return bytesconverter
...:
Your run, with FULL TRACEBACK!
In [22]: np.loadtxt(txt.splitlines(), converters={0:bytespdates2num('%Y-%m-%d')})
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-22-56c1854d614f> in <module>
----> 1 np.loadtxt(txt.splitlines(), converters={0:bytespdates2num('%Y-%m-%d')})
/usr/local/lib/python3.6/dist-packages/numpy/lib/npyio.py in loadtxt(fname, dtype, comments, delimiter, converters, skiprows, usecols, unpack, ndmin, encoding, max_rows)
1159 for x in read_data(_loadtxt_chunksize):
1160 if X is None:
-> 1161 X = np.array(x, dtype)
1162 else:
1163 nshape = list(X.shape)
TypeError: float() argument must be a string or a number, not 'datetime.datetime'
The default dtype for loadtxt is float. It has read_data (as a list of lists), and is now trying to convert it into an array, using the default dtype.
If instead I specify object as dtype:
In [23]: np.loadtxt(txt.splitlines(), converters={0:bytespdates2num('%Y-%m-%d')}, dtype=object)
...:
Out[23]:
array([datetime.datetime(2011, 1, 23, 0, 0),
datetime.datetime(2020, 3, 23, 0, 0)], dtype=object)
Or I could specify a datetime64 dtype:
In [24]: np.loadtxt(txt.splitlines(), converters={0:bytespdates2num('%Y-%m-%d')}, dtype='datetime64[D]')
Out[24]: array(['2011-01-23', '2020-03-23'], dtype='datetime64[D]')
Sorry for the caps, but I get tired asking for tracebacks and sample inputs. Providing those should be mandatory for SO questions.
With mdates (not correction in code):
In [30]: def bytespdates2num(fmt,encoding = 'utf-8'):
...: def bytesconverter(b):
...: s = b.decode(encoding)
...: return mdates.datestr2num(s)
...: return bytesconverter
...:
In [31]: np.loadtxt(txt.splitlines(), converters={0:bytespdates2num('%Y-%m-%d')})
Out[31]: array([734160., 737507.])
Evidently that's returning a number rather than a datatime object.