Unix timestamps in pandas
pandas converts whole columns at once. to_datetime with a unit reads Unix numbers; subtracting the epoch turns datetimes back into numbers.
Get the current Unix timestamp
import pandas as pd
pd.Timestamp.now(tz='UTC').timestamp() # seconds, float
pd.Timestamp.now(tz='UTC').value # nanoseconds, int
Convert a Unix timestamp to a date
df = pd.DataFrame({'ts': [1700000000, 1700003600]})
df['when'] = pd.to_datetime(df['ts'], unit='s', utc=True)
# 0 2023-11-14 22:13:20+00:00
pd.to_datetime(df['ts_ms'], unit='ms', utc=True) # milliseconds
Convert a date to a Unix timestamp
epoch = pd.Timestamp('1970-01-01', tz='UTC')
df['ts'] = (df['when'] - epoch) // pd.Timedelta('1s') # int seconds
df['ts_ms'] = (df['when'] - epoch) // pd.Timedelta('1ms')
This is the method the pandas documentation recommends. It gives the right answer whatever resolution the column uses.
Timezones
df['when_ist'] = df['when'].dt.tz_convert('Asia/Kolkata')
# naive local times: attach the zone first, then convert
df['local'] = pd.to_datetime(df['local_str']).dt.tz_localize('America/New_York').dt.tz_convert('UTC')
Common mistakes
- Without
utc=True, results are naive (no zone). Mixing naive and aware columns raises errors when you compare them. - In pandas 2 and later, datetime columns can be stored in s, ms, us or ns.
.astype('int64')returns the count in that unit, which is not always nanoseconds. Subtracting the epoch avoids the problem. tz_localizeraises on times that don't exist or happen twice at DST changes. Passnonexistent=andambiguous=to choose.
Complete example
import pandas as pd
df = pd.DataFrame({'ts': [1700000000, 1794650400]})
df['utc'] = pd.to_datetime(df['ts'], unit='s', utc=True)
df['tokyo'] = df['utc'].dt.tz_convert('Asia/Tokyo')
df['back'] = (df['utc'] - pd.Timestamp('1970-01-01', tz='UTC')) // pd.Timedelta('1s')
print(df)
Check any value with the epoch converter, or convert a whole column with the batch converter. All examples use 1700000000 (2023-11-14 22:13:20 UTC) and 1794650400 (2026-11-14 10:00:00 UTC).