Skip to content
TheChatArchiveThe Chat Archive
Qwenqwen3-235b-a22b-no-thinking· Alibaba (Qwen)Hugging Face LMArena Human Preference 140K

How do I get these plots to use scientific notation for the colorbar units? [code]

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Metallama-3.3-70b-instruct’s answer.
  1. User

    How do I get these plots to use scientific notation for the colorbar units?

        plotted_artists['DQ2PHYS'] = nn_diff_DQ2PHYS.plot(ax=axs[1,0], add_colorbar=False, cmap='RdBu_r', vmin=var_settings['DQ2PHYS']['vmin'], vmax=var_settings['DQ2PHYS']['vmax'])
        axs[1,0].set_title(f'{labels[2]} dQl/dt Bias ({var_settings["DQ2PHYS"]["unit"]})')
        axs[1,0].invert_yaxis()
        axs[1,0].set_xlabel('Latitude')
    
        plotted_artists['DQ3PHYS'] = nn_diff_DQ3PHYS.plot(ax=axs[1,1], add_colorbar=False, cmap='RdBu_r', vmin=var_settings['DQ3PHYS']['vmin'], vmax=var_settings['DQ3PHYS']['vmax'])
        axs[1,1].set_title(f'{labels[3]} dQi/dt Bias ({var_settings["DQ3PHYS"]["unit"]})')
        axs[1,1].invert_yaxis()
        axs[1,1].set_xlabel('Latitude')
    
  2. Qwen Assistant

    To display scientific notation on the colorbar for plots made with xarray's .plot(), you can manually create and customize the colorbar using matplotlib after plotting. Here's how you can do it:

    ✅ Solution: Use matplotlib.ticker to format the colorbar

    You'll need to:

    1. Plot without a colorbar (add_colorbar=False)
    2. Create a colorbar manually
    3. Apply scientific notation formatting

    Here’s how to modify your code:

    import matplotlib.pyplot as plt
    import matplotlib.ticker as ticker
    
    # For DQ2PHYS
    plotted_artists['DQ2PHYS'] = nn_diff_DQ2PHYS.plot(
        ax=axs[1,0], 
        add_colorbar=False, 
        cmap='RdBu_r', 
        vmin=var_settings['DQ2PHYS']['vmin'], 
        vmax=var_settings['DQ2PHYS']['vmax']
    )
    axs[1,0].set_title(f'{labels[2]} dQl/dt Bias ({var_settings["DQ2PHYS"]["unit"]})')
    axs[1,0].invert_yaxis()
    axs[1,0].set_xlabel('Latitude')
    
    # Add colorbar with scientific notation
    cbar2 = plt.colorbar(plotted_artists['DQ2PHYS'], ax=axs[1,0])
    cbar2.formatter = ticker.ScalarFormatter(useMathText=True)
    cbar2.formatter.set_powerlimits((0, 0))  # Forces scientific notation
    cbar2.update_ticks()
    
    # For DQ3PHYS
    plotted_artists['DQ3PHYS'] = nn_diff_DQ3PHYS.plot(
        ax=axs[1,1], 
        add_colorbar=False, 
        cmap='RdBu_r', 
        vmin=var_settings['DQ3PHYS']['vmin'], 
        vmax=var_settings['DQ3PHYS']['vmax']
    )
    axs[1,1].set_title(f'{labels[3]} dQi/dt Bias ({var_settings["DQ3PHYS"]["unit"]})')
    axs[1,1].invert_yaxis()
    axs[1,1].set_xlabel('Latitude')
    
    # Add colorbar with scientific notation
    cbar3 = plt.colorbar(plotted_artists['DQ3PHYS'], ax=axs[1,1])
    cbar3.formatter = ticker.ScalarFormatter(useMathText=True)
    cbar3.formatter.set_powerlimits((0, 0))
    cbar3.update_ticks()
    

    🔍 Explanation:

    • ticker.ScalarFormatter(useMathText=True) uses LaTeX-style scientific notation (e.g., 1e4 becomes 1×1041 \times 10^{4}).
    • set_powerlimits((0, 0)) tells matplotlib to use scientific notation for all exponents.
    • plt.colorbar(...) creates the colorbar for the given artist.

    Let me know if you'd like to apply this globally or to subplots more programmatically!

  3. User

    nn_diff_DQ2PHYS is an xarray object. Does that change your answer?

  4. Qwen Assistant

    No, the fact that nn_diff_DQ2PHYS is an xarray DataArray doesn't change the approach — in fact, it fits perfectly with what we're doing.

    ✅ Your situation:

    You're using .plot() from xarray, which returns a matplotlib QuadContourSet or imshow-like object, depending on the plot type. This object is needed to create a colorbar via plt.colorbar(), and everything I provided before works seamlessly with xarray plotting.


    🧠 Why It Still Works

    When you do:

    plotted_artists['DQ2PHYS'] = nn_diff_DQ2PHYS.plot(...)
    

    xarray internally uses matplotlib to generate the plot, and the return value (plotted_artists['DQ2PHYS']) is a valid mappable object (like a QuadContourSet or MeshQuadGrid) that can be passed to plt.colorbar().

    So this line still works fine:

    cbar2 = plt.colorbar(plotted_artists['DQ2PHYS'], ax=axs[1,0])
    

    And formatting with ticker.ScalarFormatter also applies just as well.


    ✔️ Final Code (unchanged, but confirmed safe for xarray)

    Here's the full version again, now confirmed correct for use with xarray DataArrays:

    import matplotlib.pyplot as plt
    import matplotlib.ticker as ticker
    
    # Plot DQ2PHYS
    plotted_artists['DQ2PHYS'] = nn_diff_DQ2PHYS.plot(
        ax=axs[1, 0],
        add_colorbar=False,
        cmap='RdBu_r',
        vmin=var_settings['DQ2PHYS']['vmin'],
        vmax=var_settings['DQ2PHYS']['vmax']
    )
    axs[1, 0].set_title(f'{labels[2]} dQl/dt Bias ({var_settings["DQ2PHYS"]["unit"]})')
    axs[1, 0].invert_yaxis()
    axs[1, 0].set_xlabel('Latitude')
    
    # Add scientific notation colorbar
    cbar2 = plt.colorbar(plotted_artists['DQ2PHYS'], ax=axs[1, 0])
    cbar2.formatter = ticker.ScalarFormatter(useMathText=True)
    cbar2.formatter.set_powerlimits((0, 0))
    cbar2.update_ticks()
    
    # Plot DQ3PHYS
    plotted_artists['DQ3PHYS'] = nn_diff_DQ3PHYS.plot(
        ax=axs[1, 1],
        add_colorbar=False,
        cmap='RdBu_r',
        vmin=var_settings['DQ3PHYS']['vmin'],
        vmax=var_settings['DQ3PHYS']['vmax']
    )
    axs[1, 1].set_title(f'{labels[3]} dQi/dt Bias ({var_settings["DQ3PHYS"]["unit"]})')
    axs[1, 1].invert_yaxis()
    axs[1, 1].set_xlabel('Latitude')
    
    # Add scientific notation colorbar
    cbar3 = plt.colorbar(plotted_artists['DQ3PHYS'], ax=axs[1, 1])
    cbar3.formatter = ticker.ScalarFormatter(useMathText=True)
    cbar3.formatter.set_powerlimits((0, 0))
    cbar3.update_ticks()
    

    📌 Tip: Reusable Function (Optional)

    If you’re making many such plots, consider wrapping the colorbar logic into a helper function:

    def add_sci_colorbar(artist, ax, label=None):
        cbar = plt.colorbar(artist, ax=ax)
        cbar.formatter = ticker.ScalarFormatter(useMathText=True)
        cbar.formatter.set_powerlimits((0, 0))
        cbar.update_ticks()
        if label:
            cbar.set_label(label)
        return cbar
    

    Then call:

    add_sci_colorbar(plotted_artists['DQ2PHYS'], axs[1,0])
    

    Let me know if you'd like to apply this across all subplots programmatically!

  5. User

    I tried accessing the colorbar of the xarray object and I got this error.

    ---------------------------------------------------------------------------
    AttributeError                            Traceback (most recent call last)
    Cell In[54], line 1
    ----> 1 plot_zonal_mean_tendency_bias('v6', 'squeezeformer', 'seed_7', actual_target)
    
    Cell In[53], line 57
         55 axs[1,0].set_xlabel('Latitude')
         56 cbar = plotted_artists['DQ2PHYS'].colorbar
    ---> 57 cbar.ax.yaxis.set_major_formatter(ticker.FuncFormatter(lambda x, pos: f"{x:.1e}"))
         59 plotted_artists['DQ3PHYS'] = nn_diff_DQ3PHYS.plot(ax=axs[1,1], add_colorbar=False, cmap='RdBu_r', vmin=var_settings['DQ3PHYS']['vmin'], vmax=var_settings['DQ3PHYS']['vmax'])
         60 axs[1,1].set_title(f'{labels[3]} dQi/dt Bias ({var_settings["DQ3PHYS"]["unit"]})')
    
    AttributeError: 'NoneType' object has no attribute 'ax'
    
  6. Qwen Assistant

    Thanks for sharing the error — it gives us a clear picture of what's going wrong.


    ❌ Problem Summary:

    You're trying to access the colorbar via:

    cbar = plotted_artists['DQ2PHYS'].colorbar
    

    But this fails because .colorbar is None, which causes:

    AttributeError: 'NoneType' object has no attribute 'ax'
    

    This happens because xarray does not automatically create a colorbar object unless you explicitly ask for one using add_colorbar=True. And even then, accessing it via .colorbar isn't reliable or documented behavior.


    ✅ Correct Way to Access and Format the Colorbar

    As we discussed earlier, you need to create the colorbar manually with plt.colorbar(), and then format its ticks.

    Here’s how to fix your code:


    ✅ Working Code (with Scientific Notation Formatter)

    import matplotlib.pyplot as plt
    import matplotlib.ticker as ticker
    
    # Plot DQ2PHYS
    plotted_artists['DQ2PHYS'] = nn_diff_DQ2PHYS.plot(
        ax=axs[1, 0],
        add_colorbar=False,
        cmap='RdBu_r',
        vmin=var_settings['DQ2PHYS']['vmin'],
        vmax=var_settings['DQ2PHYS']['vmax']
    )
    axs[1, 0].set_title(f'{labels[2]} dQl/dt Bias ({var_settings["DQ2PHYS"]["unit"]})')
    axs[1, 0].invert_yaxis()
    axs[1, 0].set_xlabel('Latitude')
    
    # Add and format colorbar
    cbar = plt.colorbar(plotted_artists['DQ2PHYS'], ax=axs[1, 0])
    cbar.ax.yaxis.set_major_formatter(ticker.ScalarFormatter(useMathText=True))
    cbar.ax.yaxis.get_major_formatter().set_powerlimits((0, 0))
    
    # Optional: rotate the colorbar tick labels for readability
    cbar.ax.tick_params(rotation=45)
    

    🧠 Why Your Old Code Didn't Work:

    • plotted_artists['DQ2PHYS'].colorbar doesn’t exist because add_colorbar=False.
    • Even if it did, .colorbar is not a standard or documented property.
    • Always use plt.colorbar(your_artist, ax=your_axis) to get and customize a colorbar.

    ✅ Bonus: Use Custom Formatter (if needed)

    If you want more control (e.g., custom number of decimal places), here's an example using FuncFormatter:

    def sci_notation(x, pos):
        return f"{x:.1e}"
    
    cbar.ax.yaxis.set_major_formatter(ticker.FuncFormatter(sci_notation))
    

    Let me know if you'd like to apply this formatting globally to all subplots or make it reusable in a function!

  7. User

    The colorbar in the original code gets added here:

        var_order = ['DTPHYS', 'DQ1PHYS', 'DQ2PHYS', 'DQ3PHYS', 'DUPHYS', 'DVPHYS']
        for ax, var_key in zip(axs.flat, var_order):
            img = plotted_artists[var_key]  # Use the stored artist
            if img is not None: # Check if artist exists
                cbar = fig.colorbar(img, ax=ax, orientation='vertical', fraction=0.046, pad=0.04)
            else:
                print(f"Warning: No artist found for variable {var_key} to create colorbar.")
    
  8. Qwen Assistant

    Thanks for showing the full colorbar creation loop — this helps a lot!

    You're currently adding colorbars inside a loop using fig.colorbar(...), and you want to format those colorbars to use scientific notation.


    ✅ Goal:

    Format all colorbars in your loop to use scientific notation, like:

    • 1e3 → 1×1031 \times 10^{3}
    • 0.0001 → 1×10−41 \times 10^{-4}

    ✅ Solution: Add Formatter Inside Loop

    Here's your updated code with scientific notation formatting applied to each colorbar:

    import matplotlib.ticker as ticker
    
    var_order = ['DTPHYS', 'DQ1PHYS', 'DQ2PHYS', 'DQ3PHYS', 'DUPHYS', 'DVPHYS']
    for ax, var_key in zip(axs.flat, var_order):
        img = plotted_artists[var_key]  # Use the stored artist
        if img is not None:  # Check if artist exists
            cbar = fig.colorbar(img, ax=ax, orientation='vertical', fraction=0.046, pad=0.04)
    
            # Format the colorbar ticks with scientific notation
            cbar.ax.yaxis.set_major_formatter(
                ticker.ScalarFormatter(useMathText=True)
            )
            cbar.ax.yaxis.get_major_formatter().set_powerlimits((0, 0))
            cbar.update_ticks()
    
        else:
            print(f"Warning: No artist found for variable {var_key} to create colorbar.")
    

    🧠 Explanation:

    • ticker.ScalarFormatter(useMathText=True) enables LaTeX-style scientific notation.
    • .set_powerlimits((0, 0)) ensures that everything gets formatted in scientific notation (not just small or large values).
    • cbar.update_ticks() forces the colorbar to redraw with the new format.

    📌 Optional: Rotate Labels for Readability

    If the numbers look cramped on the colorbar, you can rotate them:

    cbar.ax.tick_params(rotation=45)
    

    Or even reduce the number of ticks:

    cbar.locator = ticker.MaxNLocator(nbins=5)  # Limit to 5 ticks
    cbar.update_ticks()
    

    ✅ Final Tip: Reusable Helper Function

    If you're doing this often, wrap it up:

    def format_colorbar(cbar):
        cbar.ax.yaxis.set_major_formatter(
            ticker.ScalarFormatter(useMathText=True)
        )
        cbar.ax.yaxis.get_major_formatter().set_powerlimits((0, 0))
        cbar.update_ticks()
    

    Then call it like:

    cbar = fig.colorbar(img, ax=ax, ...)
    format_colorbar(cbar)
    

    Let me know if you'd also like to adjust label formatting, font size, or units!

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.