由于express无法实现这一点，因此我们使用graph对象为每个组绘制条形图，对于第1组和第2组，更改了通过数值确定颜色的逻辑，更改了第1组和第2组的条件，原因是如果不这样做，图例将不会进行颜色编码。

import pandas as pd
import numpy as np
import plotly.express as px 
import plotly.graph_objects as go

df = pd.DataFrame( {
    "x" : [1,2,3],
    "group1" : [np.nan, 1, -0.5],
    "group2" : [np.nan, -0.2, 1],  
}).set_index("x")

df_ = df.reset_index().melt(id_vars = 'x')

fig = go.Figure()

fig.add_trace(go.Bar(x=df_.query('variable =="group1"')['x'],
                     y=df_.query('variable =="group1"')['value'],
                     marker_color=['red' if val < 0 else 'green' for val in df_.query('variable =="group1"')['value']],
                     marker_line_color='black',
                     marker_line_width=1.5,
                     name='group1',
                     #legendgroup='group1'
                    )
             )

fig.add_trace(go.Bar(x=df_.query('variable =="group2"')['x'],
                     y=df_.query('variable =="group2"')['value'],
                     marker_color=['green' if val > 0 else 'red' for val in df_.query('variable =="group2"')['value']],
                     marker_line_color='black',
                     marker_line_width=1.5,
                     name='group2',
                     #legendgroup='group2'
                    )
             )
fig.update_layout(barmode='group', xaxis=dict(title_text='x'), yaxis=dict(title_text='value'))
fig.show()

如果您想根据x值（而不是variable组）对条形图上色，则必须单独更新轨迹（或手动绘制图形对象）：

fig = px.bar(df_, x='x', y='value', color='variable', barmode='group')
fig.for_each_trace(
    lambda trace: trace.update(marker_color=np.where(df_.loc[df_['variable'].eq(trace.name), 'value'] < 0, 'red', 'green'))
)
fig.update_layout(showlegend=False)  # Hide legend because there is no distinct group
fig.show()

输出：

为了坚持使用plotly.express，我会在 Dataframe 中添加一列，例如，df_['positive']带有布尔值，然后用这个变量给图着色。
它看起来像这样：

import pandas as pd
import numpy as np
import plotly.express as px 

df = pd.DataFrame( {
    "x" : [1,2,3],
    "group1" : [np.nan, 1, -0.5],
    "group2" : [np.nan, -0.2, 1],  
}).set_index("x")

df_ = df.reset_index().melt(id_vars = 'x')
df_['positive'] = (df_['value']>=0)

fig = px.bar(df_, x='x', y='value',barmode = 'group',
                color='positive',
                color_discrete_map={
                    True: 'green',
                    False: 'red'
                    }
)
fig.update_traces(marker_line_color='black', marker_line_width=1.5)
fig.show('browser')

其产生如下：

编辑以下评论

如果你想保持plotly.express内的颜色和组的区别，一种方法可以是添加模式...

溶液1：每个组合都有其图例条目

df = pd.DataFrame( {
    "x" : [1,2,3],
    "group1" : [np.nan, 1, -0.5],
    "group2" : [np.nan, -0.2, 1],  
}).set_index("x")

df_ = df.reset_index().melt(id_vars = 'x')
positive = (df_['value']>=0)
df_['positive'] = positive
df_['sign'] = ['positive' if x else 'negative' for x in df_['positive']]
# Each compbination of color and patterns
fig = px.bar(df_, x='x', y='value',barmode = 'group',
                color='sign',
                color_discrete_map={
                    'positive': 'green',
                    'negative': 'red'
                    },
                pattern_shape="variable")
fig.update_layout(legend_title="Groups & Signs", bargap=0.5,bargroupgap=0.1)
fig.show('browser')

其输出以下

溶液2：图例仅反映模式

# Only patterns in legend
fig = px.bar(df_, x='x', y='value', color='variable',
                barmode='group',pattern_shape="variable")
fig.update_layout(legend_title="Groups", bargap=0.5,bargroupgap=0.1)
fig.for_each_trace(
    lambda trace: trace.update(marker_color=np.where(df_.loc[df_['variable'].eq(trace.name), 'value'] < 0, 'red', 'green'))
)
fig.show('browser')

其输出：

然而，我无法从图例中"删除"绿色...