import pandas as pd
import numpy as np
from pandas import Series, DataFrame
from datetime import datetime
from datetime import timedelta

import matplotlib.pyplot as plt
%matplotlib inline

dates = [datetime(2012, 5, 1, 8, 30), datetime(2012, 5, 1, 9, 25), datetime(2012, 5, 1, 15, 30)]

dates

type(dates)

type(dates[0])

intime_dt = dates[0]
outtime_dt = dates[1]
type(intime_dt)

# The commented out lines represent attributes that ARE available in pandas Timestamp objects
#   but not in Python datetime objects
print ('Datetime: {}'.format(intime_dt))
print ('Date: {}'.format(intime_dt.date()))
print ('Month: {}'.format(intime_dt.month))
print ('Day: {}'.format(intime_dt.day))
#print ('DayOfWeek: {}'.format(intime_dt.dayofweek))
print ('Weekday: {}'.format(intime_dt.weekday()))
#print ('DayOfYear: {}'.format(intime_dt.dayofyear))
#print ('WeekOfYear: {}'.format(intime_dt.weekofyear))
#print ('Quarter: {}'.format(intime_dt.quarter))
print ('Hour: {}'.format(intime_dt.hour))
print ('Minute: {}'.format(intime_dt.minute))
print ('Second: {}'.format(intime_dt.second))
print ('Microsecond: {}'.format(intime_dt.microsecond))

los_td = outtime_dt - intime_dt
print (los_td)
type(los_td)

atts = [att for att in dir(los_td) if '__' not in att]
print(atts)

print ('Timedelta: {}'.format(los_td))
print ('Seconds: {}'.format(los_td.seconds))
print ('Total Seconds: {}'.format(los_td.total_seconds()))
print ('Microseconds: {}'.format(los_td.microseconds))
print ('Resolution: {}'.format(los_td.resolution))
print ('Min: {}'.format(los_td.min))
print ('Max: {}'.format(los_td.max))

timedelta(0,0,10)

timedelta(0,0,17).total_seconds()

atts = [att for att in dir(intime_dt) if '__' not in att]

print (atts)

np.datetime64('2012-05-01 08:30:00')

dates64 = [np.datetime64('2012-05-01 08:30:00'), np.datetime64('2012-05-01 09:25:00'), np.datetime64('2012-05-01 15:30:00')]

dates64

dt_utcnow = datetime.utcnow()
dt_utcnow

# Can create datetime64's from base Python datetimes
dt64_utcnow = np.datetime64(dt_utcnow)
dt64_utcnow

curdates64 = [np.datetime64('2014-06-01 08:30:00'), np.datetime64('2014-06-01 09:25:00'), np.datetime64('2014-06-01 15:30:00')]

curdates64

intime_dt64 = dates64[0]
outtime_dt64 = dates64[1]
print(type(intime_dt64))
print(intime_dt64)
intime_dt64

# The commented out lines represent attributes that ARE available in pandas Timestamp objects
#   but not in numpy datetime64 objects
print ('Datetime: {}'.format(str(intime_dt64)))
#print ('Date: {}'.format(intime_dt64.date()))
#print ('Month: {}'.format(intime_dt64.month))
#print ('Day: {}'.format(intime_dt.day))
#print ('DayOfWeek: {}'.format(intime_dt.dayofweek))
#print ('Weekday: {}'.format(intime_dt.weekday()))
#print ('DayOfYear: {}'.format(intime_dt.dayofyear))
#print ('WeekOfYear: {}'.format(intime_dt.weekofyear))
#print ('Quarter: {}'.format(intime_dt.quarter))
#print ('Hour: {}'.format(intime_dt.hour))
#print ('Minute: {}'.format(intime_dt.minute))
#print ('Second: {}'.format(intime_dt.second))
#print ('Microsecond: {}'.format(intime_dt.microsecond))

los_dt64 = outtime_dt64 - intime_dt64
los_dt64

# Coerce it to other units
np.timedelta64(los_dt64,'m')

ts = Series(np.random.randint(1,100,3), index=dates)

ts

ts.index

from pandas import Period, PeriodIndex

periods = PeriodIndex([Period('2012-01'), Period('2012-02'),Period('2012-03')])

ts2 = Series(np.random.randint(1,100,3), index=periods)

ts2

ts2.index

from pandas import Timestamp

Timestamp('20120501 08:30:00')

timestamps = [Timestamp('20120501 08:30:00'), Timestamp('20120501 09:25:00'), Timestamp('20120501 15:30:00')]

intime_ts = timestamps[0]
outtime_ts = timestamps[1]
type(intime_ts)

los_ptd = outtime_ts - intime_ts
print (los_ptd)
type(los_ptd)

# The pandas Timestamp data type has a number of useful attributes (dayofweek, dayofyear, weekofyear)
#   that the base Python datetime type does not.
print ('Datetime: {}'.format(intime_ts))
print ('Date: {}'.format(intime_ts.date()))
print ('Month: {}'.format(intime_ts.month))
print ('Day: {}'.format(intime_ts.day))
print ('DayOfWeek: {}'.format(intime_ts.dayofweek))
print ('Weekday: {}'.format(intime_ts.weekday()))
print ('DayOfYear: {}'.format(intime_ts.dayofyear))
print ('WeekOfYear: {}'.format(intime_ts.weekofyear))
print ('Quarter: {}'.format(intime_ts.quarter))
print ('Hour: {}'.format(intime_ts.hour))
print ('Minute: {}'.format(intime_ts.minute))
print ('Second: {}'.format(intime_ts.second))
print ('Microsecond: {}'.format(intime_ts.microsecond))

print ('Timedelta: {}'.format(los_ptd))
print ('Days: {}'.format(los_ptd.days))
print ('Seconds: {}'.format(los_ptd.seconds))
print ('Total Seconds: {}'.format(los_ptd.total_seconds()))
print ('Microseconds: {}'.format(los_ptd.microseconds))
print ('Resolution: {}'.format(los_ptd.resolution))
print ('Min: {}'.format(los_ptd.min))
print ('Max: {}'.format(los_ptd.max))

dt64 = np.datetime64('2014-06-01 08:30:00')
dt64

dt64.tolist()

ts = pd.Timestamp(dt64)
ts

dt = ts.to_datetime()
dt

np.datetime64(dt)

index = pd.date_range('2000-1-1', periods=1000, freq='M')

index

start = datetime(2014, 1, 1)
end = datetime(2014, 6, 12)
rng = pd.date_range(start, end)
rng

pd.date_range(start, end, freq='W')

data = np.random.randint(1, 100, size=163)

df = DataFrame(data, index=rng)

df.head()

# Date index slicing is easy but, BE CAREFUL, the end point is included (unlike usual Python slicing behavior).
df['2014-01-02':'2014-01-12']

df[datetime(2014,1,2):datetime(2014,1,12)]

rng2 = pd.date_range(start, end, freq='M')

len(rng2)

data[:len(rng2)]

df2 = DataFrame(data[:len(rng2)], index=rng2)

df2

from pandas.tseries.offsets import Hour, Minute
#We rarely use these but they are available

one_hour = Hour()
four_hours = Hour(4)

one_hour + four_hours

one_hour + Minute(35)

rng = pd.date_range('9/1/2012','1/1/2014',freq='M')

rng

list(rng)

rng = pd.date_range('9/1/2012','1/1/2014',freq='WOM-2TUE')

list(rng)

ts = Series(np.random.randint(1,10,4),index = pd.date_range('1/2/2013',periods=4,freq='M'))
ts

ts.shift(periods = 2)

# Common use is pct change
ts/ts.shift(1) - 1.0

# Since naive shifts leave index unchanged, some data is lost. To shift both data and index, pass in the frequency
ts.shift(1)

ts.shift(1, freq='M')

p = pd.Period('6/1/2014',freq='M')
p

p+6

# Ranges of periods is done much like ranges of dates
rng = pd.period_range('1/1/2010','6/1/2014',freq='Q')

rng

list(rng)

p.asfreq('M','start')

p.asfreq('M','end')

rng = pd.date_range('9/1/2012',periods=100,freq='D')

ts = Series(np.random.randint(1,25,len(rng)),index=rng)

ts

ts.resample('M',how='mean',kind='period')

ts.resample('M',how='sum',kind='period')

ts.resample('M',how='sum')

rng15 = pd.date_range('9/1/2012',periods=96,freq='15T')

ts15 = Series(np.random.randint(1,25,len(rng15)),index=rng15)

ts15

ts15.resample('30min',how='mean',closed='right',label='left')

ts30 = ts15.resample('30min',how='mean',closed='left',label='left')
ts30

ts60 = ts15.resample('60min',how='ohlc',closed='left',label='left')
ts60

ts30

ts30.resample('15min')

close_px_all = pd.read_csv('stock_px.csv', parse_dates=True, index_col=0)

close_px_all.head()

close_px = close_px_all[['AAPL','MSFT','XOM']]

close_px[1:10]

close_px['AAPL'].plot()

close_px.plot()

# And here's what it looks like if we just plot one year. Notice the auto x-axis formatting.
close_px.ix['2009'].plot()

close_px.index

list(close_px.index)[1:25]

close_px['AAPL'].ix['01-2011':'03-2011'].plot()

appl_q = close_px['AAPL'].resample('Q-DEC')

appl_q

appl_q.ix['2009':].plot()

close_px.AAPL.plot()
pd.rolling_mean(close_px.AAPL,250).plot()

pd.rolling_mean(close_px.AAPL,250)

pd.rolling_mean(close_px.AAPL,250,min_periods=10)

close_px.AAPL.plot()
pd.rolling_mean(close_px.AAPL,250,min_periods=10).plot()

# Things like rolling standard deviation are also possible
appl_std250 = pd.rolling_std(close_px.AAPL,250,min_periods=10)

appl_std250.plot()