Having troubles converting pandas datetime to unix time stamp

Question

What I need to do is convert 'year-month-day' timestamp to Unix time stamp do somethings with it then change it back to date time series. I am working with '1999-09-07' as my timestamp. I am getting an error : invalid literal for int() with base 10: '1999-09-07'

df1['timestamp'] = df1['timestamp'].astype(np.int64) // 10**9
#Got back this 
ERROR:invalid literal for int() with base 10: '1999-09-07'


df1 = pd.read_csv('stock_CSV/' + ticker + '.csv')
pd.to_datetime(df1['timestamp'],unit='ns', origin='unix')
df1['timestamp'] = df1['timestamp'].astype(np.int64) // 10**9
#
#....some code
#
pd.to_datetime(df1['timestamp'], unit='s')

What I am expecting is a my dates converted to unix timestamp then converted back

Your timestamp looks like string. Try `df1['timestamp'] = pd.to_datetime(df1['timestamp'])` first. — Quang Hoang, Sep 13 '19 at 17:33
`df1['timestamp'] = pd.to_datetime(df1['timestamp']).astype...` — Quang Hoang, Sep 13 '19 at 17:42
I don't think I can minimalize it even more than what I have here. import the data, convert to datetime then try to convert it to unix timestamp, then convert it back — David Long, Sep 13 '19 at 18:32
Thank you Quang Hoang I had to while importing it to set index and parse dates then use what you have here to get it to work — David Long, Sep 18 '19 at 19:24

score 0 · Answer 1 · answered Sep 14 '19 at 17:02

0

Calling astype('int64') on a Timestamp Series return the number of seconds even though the Timestamps may have a resolution of nanosecond.

You can do it the old way, by counting the seconds from Jan 1, 1970 to the timestamps:

# random epoch times, in nanoseconds
t = np.random.randint(1e9 * 1e9, 2e9 * 1e9, 10)

# convert to Timestamps
dates = pd.to_datetime(t, unit='ns')

# convert back to nanoseconds
epoch = (dates - pd.Timestamp('1970-01-01')).total_seconds() * 1e9

# verify that we did the conversion correctly
assert np.isclose(t, epoch).all()

answered Sep 14 '19 at 17:02

Code Different

90,614
16
144
163

I am not sure with what you have here to put it together for my data. You have random times as t then converting them to date-time then subtracting Jan, 1 1970, maybe would I or how would I convert my timestamps from 'year-month-day' to seconds? – David Long Sep 16 '19 at 18:46
The Unix timestamp is simply the number of seconds since midnight on Jan 1, 1970. To get the nanoseconds, you multiply that by 1e9 – Code Different Sep 16 '19 at 18:47

Having troubles converting pandas datetime to unix time stamp

1 Answers1