import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from datetime import timedelta
[docs]
def generate_sample_data():
freq = "15T"
start = pd.Timestamp("2010-01-01")
end = pd.Timestamp("2012-01-01")
index = pd.date_range(start, end, freq=freq)
n = len(index)
cols = [f"Series{i+1}" for i in range(12)]
data = np.random.randn(n, 12)
df = pd.DataFrame(data, index=index, columns=cols)
rng = np.random.default_rng(seed=0)
for col in df.columns:
for _ in range(20):
start_idx = rng.integers(0, n - 1)
gap_length = rng.integers(1, 201)
end_idx = min(start_idx + gap_length, n)
df.loc[df.index[start_idx:end_idx], col] = np.nan
gap_end_candidates = rng.choice(df.columns, size=6, replace=False)
intervals_per_day = int((24 * 60) / 15)
for col in gap_end_candidates:
gap_length = rng.integers(1, 2 * intervals_per_day + 1)
df.loc[df.index[-gap_length:], col] = np.nan
df.loc[df.index < (start + pd.DateOffset(years=1)), df.columns[0]] = np.nan
return df
[docs]
def plot_missing_data(df, ax, min_gap_duration, overall_start, overall_end):
overall_start_num = mdates.date2num(overall_start)
overall_end_num = mdates.date2num(overall_end)
overall_color = "skyblue"
gap_color = "orange"
boundary_gap_color = "darkorange"
bar_height = 0.8
ax.cla()
y_ticks = []
y_labels = []
for i, col in enumerate(df.columns):
# Full-range background bar
ax.broken_barh(
[(overall_start_num, overall_end_num - overall_start_num)],
(i - bar_height / 2, bar_height),
facecolors=overall_color,
alpha=0.6,
)
series = df[col]
mask = series.isna()
if mask.any():
groups = (mask != mask.shift()).cumsum()
for _, group in mask.groupby(groups):
if group.iloc[0]: # missing segment
gs = group.index[0]
ge = group.index[-1] + pd.Timedelta(minutes=15)
actual = ge - gs
if actual < min_gap_duration:
extra = min_gap_duration - actual
gs = max(gs - extra / 2, overall_start)
ge = min(ge + extra / 2, overall_end)
color = (
boundary_gap_color
if (gs <= overall_start or ge >= overall_end)
else gap_color
)
ax.broken_barh(
[
(
mdates.date2num(gs),
mdates.date2num(ge) - mdates.date2num(gs),
)
],
(i - bar_height / 2, bar_height),
facecolors=color,
)
y_ticks.append(i)
# Missing percentage label
perc = series.isna().mean() * 100
if perc == 0:
lbl = f"{col} (0%)"
elif perc < 0.01:
lbl = f"{col} (<0.01%)"
else:
lbl = f"{col} ({perc:.2f}%)"
y_labels.append(lbl)
ax.set_yticks(y_ticks)
ax.set_yticklabels(y_labels)
ax.set_xlabel("Time")
ax.set_title(f"Missing Data (Min gap = {min_gap_duration})")
ax.xaxis_date()
ax.figure.autofmt_xdate()
# Fix y-limits so zooming on x-axis only
ax.set_ylim(-0.5, len(df.columns) - 0.5)
[docs]
def interactive_gap_plot(df):
overall_start = df.index[0]
overall_end = df.index[-1] + pd.Timedelta(minutes=15)
fig, ax = plt.subplots(figsize=(12, 6))
prev_xlim = [None, None]
# Initial draw
default_gap = timedelta(days=4)
plot_missing_data(df, ax, default_gap, overall_start, overall_end)
def on_draw(event):
# Preserve user zoom x-limits
x0, x1 = ax.get_xlim()
if prev_xlim[0] == x0 and prev_xlim[1] == x1:
return
prev_xlim[0], prev_xlim[1] = x0, x1
# Determine new min-gap based on visible span
d0 = mdates.num2date(x0)
d1 = mdates.num2date(x1)
years_view = (d1 - d0).total_seconds() / (365.25 * 24 * 3600)
if years_view >= 12:
mg = timedelta(days=12)
elif years_view >= 8:
mg = timedelta(days=4)
elif years_view >= 4:
mg = timedelta(hours=18)
elif years_view >= 1:
mg = timedelta(hours=4)
else:
mg = timedelta(hours=1)
# Redraw bars then restore zoom
plot_missing_data(df, ax, mg, overall_start, overall_end)
ax.set_xlim(x0, x1)
# Use draw_event for reliable detection after toolbar zoom/pan
fig.canvas.mpl_connect("draw_event", on_draw)
plt.tight_layout()
plt.show()
if __name__ == "__main__":
df = generate_sample_data()
interactive_plot(df)
