Plotly 使用 groupby 的置信区间折线图

Question

I'd like to plot time series simulation data as a mean with confidence intervals and compare multiple scenarios. Using the pandas groupby() and agg() functions a calculate the mean and confidence interval (upper and lower limit) see sample data (the actual data can be retrieved from my github .

Edit: I added the raw data to the and the code (as a jupyter notebook) to the git

Plotting this data for one specific parameter combination (selecting data via df = df.loc[(slice(None),1, True)] ) seams simple enough:

myFig = go.Figure([
go.Scatter(
    name='Mittelwert',
    #x=df['tick'],
    y=df['mean'],
    mode='lines',
    line=dict(color='rgb(31, 119, 255)'),
),
go.Scatter(
    name='Konfidenzintervall',
    #x=df['tick'],
    y=df['ci95_hi'],
    mode='lines',
    marker=dict(color="#644"),
    line=dict(width=0),
    showlegend=True
),
go.Scatter(
    name='Konfidenzintervall',
    #x=df['tick'],
    y=df['ci95_lo'],
    marker=dict(color="#448"),
    line=dict(width=0),
    mode='lines',
    fillcolor='rgba(130, 68, 68, 0.5)',
    fill='tonexty',
    showlegend=True
)
])
myFig.update_layout(
    xaxis_title='X Achse',
    yaxis_title='Y Achse',
    title='Continuous, variable value error bars',
    hovermode="x"
    )
myFig.show()

This code gives me the that beautiful plot. The issue is a do not know how to properly plot the grouped data. When I don't select a subset all data is plotted at once. Therefore i tried to use color , facet_col and facet_row which I could get working using

px.line(reset_df, x = "tick", y="mean", color="first_factor",facet_col="second_factor")

(Since plotly apparently can't handle MultiIndex Dataframe i used reset_index() to get a DataFrame with a 'RangeIndex' first). The issue is with the latter approach I'm now missing the confidence interval and don't know how to add it ( c.f. this plot ).

How can I have both the ci and the grouped data within one plot? If this is not possible with pandas is it with bokeh? Thank you very much

Answer 1

( this is work in progress )

---

It's still not 100% clear to me how you'd like to display your data here. In your code sample:

px.line(reset_df, x = "tick", y="mean", color="first_factor",facet_col="second_factor")

... you're using first_factor and second_factor without saying what those are. And judging by the column names in your dataframe it could be almost any of:

['mean', 'median', 'count', 'std', 'min', 'max', 'ci95_lo', 'ci95_hi']

..or your index:

'(1.0, 0.0, False)'

But since it seems clear that you'd like to display a confidence interval around a mean across some categories, I'm guessing that this is what you're looking for:

Plot 1:

Plot 2: Zoomed in on first plot

The way to get there is full of pit-falls, and I'm not going to waste anyone's time talking about the details if this is not what you're looking for. But if it is , I'd be happy to explain everything.

Complete code:

df = pd.read_json("https://raw.githubusercontent.com/thefeinkoster/plotly_issue/main/aggregatedData.json")
df = df.reset_index()
df = pd.concat([df, df['index'].str.split(',', expand = True)], axis = 1)
df = df.rename(columns = {0:'ix', 1:'scenario', 'ci95_lo':'lo', 'ci95_hi':'hi'})

df.ix = [int(float(i[1:])) for i in df.ix]
df = df[df[2]==' True)']

dfp=pd.melt(df, id_vars=['ix', 'scenario'], value_vars=df.columns[-12:])
dfp = dfp[dfp['variable'].isin(['ix', 'mean', 'lo', 'hi'])] 

dfp = dfp.sort_values(['ix', 'variable'], ascending = False)

fig = px.line(dfp, x = 'ix', y = 'value', color = 'variable', facet_row = 'scenario')
fig.update_layout(height = 800)

fig.update_traces(name = 'interval', selector = dict(name = 'hi'))
fig.update_traces(fill = 'tonexty')
fig.update_traces(fillcolor = 'rgba(0,0,0,0)', selector = dict(name = 'mean'))
fig.update_traces(fillcolor = 'rgba(0,0,0,0)', line_color = 'rgba(44, 160, 44, 0.5)',
                  showlegend = False, selector = dict(name = 'lo'))

fig.update_yaxes(range=[0.44, 0.54])
f = fig.full_figure_for_development(warn=False)
fig.show()