.. _python-outputs: # Creating custom model outputs with Python This guide gives an example of using a python script to customize the output files produced by a RiskScape model. For example, python packages like `matplotlib` can be great for taking CSV results and plotting them as a graph for better visualization. RiskScape lets you integrate this as part of the model run process, so that plots and other outputs can be generated automatically any time the model is run. ## Before we start This guide is aimed at users who already have some familiarity with Python and want to use it to create custom model outputs, such as PDFs, graphs or map plots. We expect that you: - Have been through the :ref:`getting-started` tutorial. - Have some basic Python knowledge and have Python installed on your computer. .. note:: There are several different approaches to integrating RiskScape with Python. This tutorial is specifically looking at using Python to change how model results are *presented*. For other ways to integrate Python with RiskScape, see :ref:`python_disambiguation`. ## Getting started ### Setup Click [here](../python-output.zip) to download the example project we will use in this guide. Unzip the file into the :ref:`top_level_dir` where you normally keep your RiskScape projects. This project contains a working example of the `building-damage` model from the :ref:`getting-started` guide. ### CPython The default implementation of the Python language that most people use is technically called [CPython](https://en.wikipedia.org/wiki/CPython) (it's written in C, hence the name). By executing a CPython 'post-processing' script as part of your model run, it gives you a lot of flexibility in tailoring outputs to suit your risk modelling needs. .. note:: In order to use a Python post-processing script, you need to have configured RiskScape to use :ref:`cpython-impl`. We use a few Python libraries in this tutorial. If you want to follow along, you'll need to have installed: - `pandas` - `geopandas` - `shapely` - `matplotlib` - `markdown_pdf` - `tabulate` Each library can be installed by running `pip install `, or by using your system package manager. See [python.org](https://packaging.python.org/en/latest/tutorials/installing-packages) for more help installing libraries. ### The building damage model Firstly, try running the building damage model 'as is' by entering the following command into your terminal: ``` riskscape model run building-damage ``` .. tip:: If the RiskScape command produced an error, try checking that ``riskscape -V`` runs OK, that the current working directory is where you unzipped the example, and that RiskScape is setup correctly to use CPython. Open the `project.ini` file in a text editor. This sets up the the building-damage model for RiskScape to use, including parameters it supports and a Python function to determine the damage to each building from a tsunami event. This tutorial will look at how we can update the model in this project to support a post-processing script that can transform the model's parameters and outputs into more custom model outputs. The project directory contains several Python files (`plot.py`, `choropleth.py`, and `pdf.py`) that will do the Python processing work for us. ### Bar graph We will start off by using Python to create a simple bar graph. Open the `plot.py` file in your text editor. It should look like the following: ```python import matplotlib.pyplot as plt import pandas as pd def bar_graph(df, filename): # bar graph plot states = ['Light', 'Minor', 'Moderate', 'Severe', 'Collapsed'] total_count = [ sum([ region for region in df[state] ]) for state in states ] plt.bar(states, total_count) plt.title('Number of damaged buildings') # also add the total count as a label for i, y in enumerate(total_count): plt.text(i, y, y, ha='center') plt.savefig(filename) # RiskScape post-processing-script entry point: def function(metadata): # RiskScape passes us the output filepaths as a Python dictionary outputs = metadata['outputs'] # open the summary.csv file as a pandas dataframe df = pd.read_csv(outputs['summary']) plot_filename = model_output('building-damage-states.png') # use matplotlib to turn the dataframe into a bar graph bar_graph(df, plot_filename) return plot_filename ``` This file contains two Python functions: - the first function `bar_graph()` uses [Matplotlib](https://matplotlib.org/) code to turn a [Pandas](https://pandas.pydata.org/) DataFrame into a simple bar graph. - the second `function()` is the entry-point for the post-processing script that RiskScape calls. It reads the `summary.csv` output file produced by RiskScape into a Pandas DataFrame, and then calls the Matplotlib plotting code. The last step is to tell RiskScape to execute this script once the model has run. We can do this by adding a `post-processing-script` parameter to the model's INI configuration. Open up the `project.ini` file and uncomment the last line in the file so that the model definition looks like this: ```ini [model building-damage] description = Model that calculates building damage framework = pipeline location = building-damage-pipeline.txt # Call this python script once the pipeline has run post-processing-script = plot.py ``` Save the file and enter the following command to run the model again: ``` riskscape model run building-damage ``` You should now see an extra line in the list of outputs for a `building-damage-states.png` file. If you open that file up, you should see something like this. .. image:: ../screenshots/building-damage-states.png :target: ../_images/building-damage-states.png :alt: A bar chart showing the number of buildings in each damage state (Light, Minor, Moderate, Severe, and Collapse) .. note:: This model uses a weighted random choice to assign a damage state to each building, so you will get a slightly different result every time you run the model, due to randomness. RiskScape does support *reproducible* randomness by using a random-seed, such as a feature ``ID`` attribute. ### Registering model outputs One important thing that `plot.py` does happens on the following line: ```python plot_filename = model_output('building-damage-states.png') ``` The `model_output()` function is a special Python function provided by RiskScape. It tells RiskScape that our Python function is writing an output file (called `building-damage-states.png`). RiskScape will then make sure the `building-damage-states.png` file gets saved in the same directory as all the other model outputs. The `model_output()` function simply returns the text string it is given, so it should be compatible to use with any Python code where you need to create a file. .. tip:: We recommend using ``model_output()`` whenever you save a file in Python code. If you don't, the file will still get written but it might end up in a different output directory, and the model won't be compatible with the RiskScape Platform. .. note:: If you split up your Python code across multiple ``.py`` files, the ``model_output()`` function will not be accessible from other ``.py`` files you might import. RiskScape only adds it to the global name-space of your ``post-processing-script`` by default. ### Choropleth map Next, let's add an output that shows our data on a map. Go back to your `project.ini` file and change the `post-processing-script` line to use `choropleth.py`. It should look like this: ```python [model building-damage] description = Model that calculates building damage framework = pipeline location = building-damage-pipeline.txt # Call this python script once the pipeline has run post-processing-script = choropleth.py ``` Save the file and run the model again. The model should now produce a new `regional-collapsed-buildings.png` output. It should look something like this: .. image:: ../screenshots/regional-collapsed-buildings.png :target: ../_images/regional-collapsed-buildings.png :alt: A map of Upolu, with different regions coloured by the number of collapsed buildings. Open the `choropleth.py` file in your text editor. It should look like this: ```python import matplotlib.pyplot as plt import geopandas as gpd def choropleth_map(gdf, filename, column): ax = gdf.plot(column=column, cmap='Reds', legend=True) # labels gdf.apply(lambda x: ax.annotate(text=x['Region'], xy=x.geometry.centroid.coords[0], ha='center', size=5), axis=1) ax.set_axis_off() ax.set_title(column + ' building damage in Upolu by region') plt.savefig(filename) return filename # RiskScape post-processing-script entry point: def function(metadata): # RiskScape passes us the output filepaths as a Python dictionary outputs = metadata['outputs'] # open the regional-impact.geojson file as a geopandas dataframe df = gpd.read_file(outputs['regional-impact']) # get the value specified for the choropleth model parameter. # This lets us control which column in the output to plot dynamically column = metadata['parameters']['choropleth'] # we can even change the file that gets produced dynamically to match the column name output_filename = model_output('regional-%s-buildings.png' % column.lower()) # use matplotlib to turn the dataframe into a choropleth plot choropleth_map(df, output_filename, column) return output_filename ``` This is the Python code that was used to produce the choropleth map. It follows a similar pattern to our earlier `plot.py` code - it accepts the DataFrame as well as the filename it should write the image out to, but it uses the `plot` method from GeoPandas to succinctly plot the spatial data, rather than plotting a bar graph with Matplotlib. This Python file also has some additional code that dynamically changes the data column that is being plotted in the map, which we'll look at in more detail next. ### Python parameters The metadata passed to the Python post-processing script also contains any parameters that were used by the RiskScape model. This choropleth example uses a model parameter (called `choropleth`) to decide which attribute (or data column name) to use for colouring the map. The `choropleth` parameter defaults to 'Collapsed', but we can change it when running the model via a command line parameter: ``` riskscape model run building-damage -p choropleth=Exposed ``` The `choropleth.py` Python code will now plot the `Exposed` attribute instead of the default `Collapsed` attribute. This also changes the name of the output file to reflect the data it contains. Open the new `regional-exposed-buildings.png` output and see how it compares to the `regional-collapsed-buildings.png` output produced previously. Try running the following commands to plot the total damaged buildings by region, and the mean inundation depth (for exposed buildings). Open the plots that are produced and see how they have changed. ``` riskscape model run building-damage -p choropleth=Damaged riskscape model run building-damage -p choropleth=Mean_Depth ``` ### PDF outputs You can also use Python to generate PDF outputs. This is useful for generating reports, or even just to collect all your other outputs on a single page for sharing. Update your `project.ini` file so that `pdf.py` is used as the post-processing script: ```python [model building-damage] description = Model that calculates building damage framework = pipeline location = building-damage-pipeline.txt # Call this python script once the pipeline has run post-processing-script = pdf.py ``` The `pdf.py` code uses a Python library called MarkdownPDF to convert [markdown](https://en.wikipedia.org/wiki/Markdown) text into a PDF document. Markdown is a simple way to apply styling, such as headings and formatting, to a plain-text document. .. note:: There are many different Python libraries that you can use to create a PDF. We have used MarkdownPDF here because it works well as a simple example. Save your `project.ini` file and run the model again. It should now produce a `Report-Summary.pdf` PDF output. Open the PDF file - it should look something like this: .. image:: ../screenshots/pdf-output.png :target: ../_images/pdf-output.png :alt: The first section of a PDF report. It includes a header, introductory paragraph, and the bar chart from above. The template markdown for our report is stored in a file called `template.md`. Open the file in your preferred text editor and have a look. You'll notice some text in curly brackets (braces). The Python `pdf.py` code is reading this template file and then swapping out that placeholder text for our actual model results. This next part is the Python code that reads the `template.md` file and replaces the placeholder values (in `{}`s) with the actual results coming out of the model. ```python # replace the {placeholder} values in the template with the actual results with open("template.md") as template: text = template.read().format( total_damaged=total_damaged, total_buildings=sum(totals.values()), total_collapsed=totals['Collapsed'], bar_graph_fname=bar_graph_filename, choropleth_map_fname=choropleth_map_filename ) ``` The next bit appends a table of the regional results to the PDF. Handily, the Pandas DataFrame has a convenient `.to_markdown()` method, so we don't have to make the table ourselves. ```python # insert the simplified table of results text += "\n" + table.to_markdown(index=False) ``` Finally, the code passes the markdown string to MarkdownPDF for it to generate our PDF. ```python pdf = MarkdownPdf() pdf.add_section(Section(text), user_css=style) pdf.save(model_output('Report-Summary.pdf')) ``` #### Optional PDF Styling We skipped over one part of the `pdf.py` code, which applies styling to the final PDF: ```python with open("style.css") as file: style = file.read() ``` This is applies [CSS](https://developer.mozilla.org/en-US/docs/Web/CSS) to the PDF, which is the same styling used by web pages. In this case, it changes the colour and font of the heading, and applies styling to the table. With the approach used in this example, markdown supports basic font styling (such as bold and italics), whereas CSS would be used to change other aspects (such as the font type, size, and colour). Another alternative approach would be to use [LaTeX](https://www.latex-project.org/about/) to control the styling when generating a PDF from Python. ## Testing your Python code ### Running Python manually When you're writing your own Python code, it can take quite a long time to test if you have to run your entire RiskScape model every time. You can test your Python code manually, outside of the RiskScape model, by using the ``if __name__ == '__main__':`` `Python idiom `_. The bottom of your `pdf.py` script already contains the following code: ``` # code to manually run the script outside of RiskScape if __name__ == '__main__': metadata = { 'parameters': {'choropleth': 'Collapsed'}, 'outputs': { 'regional-impact': './output/example/regional-impact.geojson', 'summary': './output/example/summary.csv', } } def model_output(name): filepath = "output/example/%s" % name print('Writing ' + filepath) return filepath function(metadata) ``` This mimics the way RiskScape calls your function after the pipeline completes, by manually setting up a `metadata` dictionary containing the parameters and outputs to use. To make sure there's an output for it to use, run the model using the `--output` parameter to write the model results to a fixed location that matches what the script expects: ``` riskscape model run building-damage --replace --output output/example ``` Then try running the `pdf.py` script manually with python: ``` python pdf.py ``` The Python script should have generated `.png` and `.pdf` files in the `output/example` directory, similar to what happens in a RiskScape model run. .. tip:: Running the Python script manually allows you to test changes to your Python code without having to run the *entire* RiskScape model every time. ### Input data subset Another alternative to speed up testing your Python code is to simply run the model over a subset of results. Instead of running the model over the entire building dataset, you could limit the model run to the first 50 buildings. This particular model has a `input_exposures_rows` parameter that controls the number of rows of input data that the model loads. Try running the following command: ``` riskscape model run building-damage -p input_exposures_rows=50 ``` Now if you run the model again, only the first 50 buildings will be included in the results. Although it does not make a huge difference to the model run-time in this simple example, it can make a big difference if your model is processing millions of assets. Though you obviously won't get an accurate result without all the data, the model will run much quicker, meaning you can test changes to your Python code faster. This is useful when you are changing the format of the model results as well as the Python code, e.g. you realized you needed to add an extra column in the results CSV. .. note:: Advanced users could look at the ``building-damage-pipeline.txt`` code to see how the ``$input_exposures_rows`` parameter is used in the pipeline. ## Summary This tutorial has covered some simple examples of how you can use Python to create customized outputs, such as plots, maps, or even PDF reports, when you run a RiskScape model. Python is a very flexible language, so potentially anything you could do in Python could be integrated with the RiskScape model run.