Plotly: bar plot with color red<0, green>0, divided by groups

Given a dataframe with 2 groups: (group1, group2), that have values > and < than 0: plot:

• Bar plot
• x = x
• y = values, divided by group1, group2
• color = red if value<0, green if value>0
• legend shows group1, grou2 with different colors.

My current code however is not coloring as i would expect, and the legend is shown with the same color:

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')
fig = px.bar(df_, x='x', y='value', color='variable', barmode='group')
fig.update_traces(marker_color=['red' if val < 0 else 'green' for val in df_['value']], marker_line_color='black', marker_line_width=1.5)
fig.show()

OUT with indications of what i want to achieve:

To stick with plotly.express, I would add a column to your dataframe, e.g. df_["positive"] with a boolean, and then color your plot by this variable.

It would look like this:

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")

which yields the following :

EDIT following comments

If you want to keep the colors AND the group distinction within plotly.express, one way could be to add patterns...

Solution 1 : Every combination has its legend entry

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")

which outputs the following

Solution 2 : Legend only reflects patterns

# 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")

which outputs : However I was not able to 'remove' the green color from the legend...