Boxplot placed on time axis

Question

I want to place a series of (matplotlib) boxplots in a time axis. They are series of measurements taken on different days along a year. The dates are not evenly distributed and I am interested on the variation along time.

---

Easy version

I have a pandas DataFrame with indexes and series of numbers, more or less like this: (notice the indexes):

np.random.seed(12345)
data = np.array( [ np.random.normal( i, 1, 10 ) for i in range(3) ] )
ii = np.array([ 3, 5, 8 ] )
df = pd.DataFrame( data=data, index=ii )

For each index, I need to make a boxplot, which is no problem:

plt.boxplot( [ df.loc[i] for i in df.index ], vert=True, positions=ii )

Time version

The problem is, I need to place the boxes in a time axis, i.e. place the boxes on a concrete date

np.random.seed(12345)
data = np.array( [ np.random.normal( i, 1, 10 ) for i in range(3) ] )
dates = pd.to_datetime( [ '2015-06-01', '2015-06-15', '2015-08-30' ] )
df = pd.DataFrame( data=data, index=dates )
plt.boxplot( [ df.loc[i] for i in df.index ], vert=True )

However, if I incorporate the positions:

ax.boxplot( [ df.loc[i] for i in df.index ], vert=True, positions=dates )

I get an error:

TypeError: Cannot compare type 'Timedelta' with type 'float'

A look up on the docs shows:

plt.boxplot?

positions : array-like, default = [1, 2, ..., n]

Sets the positions of the boxes. The ticks and limits are automatically set to match the positions.

---

Wished time version

This code is intended to clarify, narrow down the problem. The boxes should apppear there, where the blue points are placed in the next figure.

np.random.seed(12345)
data = np.array( [ np.random.normal( i, 1, 10 ) for i in range(3) ] )
dates = pd.to_datetime( [ '2015-06-01', '2015-06-15', '2015-08-30' ] )
df = pd.DataFrame( data=data, index=dates )

fig, ax = plt.subplots( figsize=(10,5) )
x1 = pd.to_datetime( '2015-05-01' )
x2 = pd.to_datetime( '2015-09-30' )
ax.set_xlim( [ x1, x2 ] )

# ax.boxplot( [ df.loc[i] for i in df.index ], vert=True ) # Does not throw error, but plots nothing (out of range)
# ax.boxplot( [ df.loc[i] for i in df.index ], vert=True, positions=dates ) # This is what I'd like (throws TypeError)

ax.plot( dates, [ df.loc[i].mean() for i in df.index ], 'o' )  # Added to clarify the positions I aim for

---

Is there a method to place boxplots in a time axis?

---

I am using:

python: 3.4.3 + numpy: 1.11.0 + pandas: 0.18.0 + matplotlib: 1.5.1

Answer 1

So far, my best solution is to convert the units of the axis into a suitable int unit and plot everything accordingly. In my case, those are days.

np.random.seed(12345)
data = np.array( [ np.random.normal( i, 1, 10 ) for i in range(3) ] )
dates = pd.to_datetime( [ '2015-06-01', '2015-06-15', '2015-08-30' ] )
df = pd.DataFrame( data=data, index=dates )

fig, ax = plt.subplots( figsize=(10,5) )
x1 = pd.to_datetime( '2015-05-01' )
x2 = pd.to_datetime( '2015-09-30' )
pos = ( dates - x1 ).days

ax.boxplot( [ df.loc[i] for i in df.index ], vert=True, positions=pos )
ax.plot( pos, [ df.loc[i].mean() for i in df.index ], 'o' )

ax.set_xlim( [ 0, (x2-x1).days ] )
ax.set_xticklabels( dates.date, rotation=45 )

The boxplots are placed on their correct position, but the code seems a bit cumbersome to me.

More importantly: The units of the x-axis are not "time" anymore.