3

I got a very similar question to this one : Pandas graphing a timeseries, with vertical lines at selected dates but the solution doesn't works with Timedelta.

Consider this series:

In:
avg_hr.head()

Out:
00:00:00     69.000000
00:00:01     93.750000
00:00:02     93.125000
00:00:03     92.900000
00:00:04     93.222222
00:00:05     93.222222
...
Name: bpm, Length: 253, dtype: float64

I can select element in this series like this:

In:
avg_hr[pd.Timedelta(seconds=3)]

Out:
92.9

I can generate a graph like this:

In:
avg_hr.plot()

avg_hr.plot()

But, I can't plot vertical lines with TimeDelta like this:

In:
plt.axvline(x=pd.Timedelta(seconds=110), color='r', linestyle='dashed', linewidth=2)

Out:
TypeError: Cannot compare type 'Timedelta' with type 'float64'

Though, if I use a float or int, the vertical lines appear at position 0.

In:
plt.axvline(x=110, color='r', linestyle='dashed', linewidth=2)

avg_hr.plot()

How can I plot vertical lines using this timedelta index?

EDIT:

Even if I use directly the keys used on x-axis, I got the same error: 

In:
for key in avg_hr.keys():
    ax.axvline(x=key, color='r', linestyle='dashed', linewidth=2)

Out:
TypeError: Cannot compare type 'Timedelta' with type 'float64'
probitaille
  • 1,899
  • 1
  • 19
  • 37

2 Answers2

10

I figured out that even if I work in seconds, and that the axis label show the time in second, it's in fact in nanoseconds!

From the documentation Pandas Time Deltas:

Pandas represents Timedeltas in nanosecond resolution using 64 bit integers

So, in the example of my question, when I called this, the vertical line was not at position 0, but in fact at position 110 nanoseconds (so very close to 0 with this scale):

plt.axvline(x=110, color='r', linestyle='dashed', linewidth=2)

The solution is simply to convert your x value in nanoseconds:

x_ns = pd.Timedelta(seconds=110) / pd.Timedelta(1,'ns') #Seconds to nanoseconds
plt.axvline(x=x_ns, color='r', linestyle='dashed', linewidth=2)

I found this when I tried to change the xlim, then I saw that everything was scale to nanoseconds. So, the same conversion needed to be applied to xlim.

ax1.set_xlim([0, 110])

enter image description here

The result with multiple verticals lines

enter image description here

Done with: 

#Add verticals lines for specific event
plt.axvline(x=pd.Timedelta(seconds=120) / pd.Timedelta(1,'ns'), color='r', linestyle='dashed', linewidth=2)
plt.axvline(x=pd.Timedelta(seconds=185) / pd.Timedelta(1, 'ns'), color='r', linestyle='dashed', linewidth=2)
plt.axvline(x=pd.Timedelta(seconds=210) / pd.Timedelta(1, 'ns'), color='r', linestyle='dashed', linewidth=2)
plt.axvline(x=pd.Timedelta(seconds=225) / pd.Timedelta(1, 'ns'), color='r', linestyle='dashed', linewidth=2)
probitaille
  • 1,899
  • 1
  • 19
  • 37
  • 1
    Really neat, you could instead, reformatted your dataframe, it seems that `avg_hr = pd.DataFrame({'data':pd.Series(np.random.randn(240)).cumsum(),'time': pd.to_timedelta(np.arange(240), unit='s')})` ; `avg_hr.plot()` ; `plt.axvline(x=110,color='r', linestyle='dashed', linewidth=2)` ; `plt.show()` works good. – Vinícius Figueiredo Jul 21 '17 at 19:35
0

I have encountered a similar problem. The solution for the TimeDelta index I used was the total_seconds property which returns float in seconds. ("Total duration of timedelta in seconds (to ns precision)")

So,

plt.axvline(pd.Timedelta(seconds=120).total_seconds, color='r', linestyle='dashed', linewidth=2)

should do the trick.

badluck
  • 183
  • 1
  • 11