.. _expression_tutorial: # How to write RiskScape expressions RiskScape expressions are an integral part of model pipelines. :ref:`expressions` explains in more detail what expressions are, whereas this page will look at practical examples. We will start off with the basics, and gradually cover some of the more nuanced aspects of expressions. ## Before we start This tutorial is aimed at people who already have experience running RiskScape models, but want more flexibility in how they can customize their models and the results they produce. We expect that: - You are familiar with using a command line in either Linux or Windows. - You feel comfortable doing some basic coding, such as in Python, R, or SQL. Or you are proficient with using spreadsheet formula and have a willingness to learn. - You have RiskScape installed on your system. The aim of this tutorial is to give you enough knowledge to take a pipeline generated by the wizard and start modifying some of the expressions in it yourself. ## Setup Click [here](../expressions-tutorial.zip) to download the example project we will use in this guide. Unzip the file into a new directory, e.g. `C:\RiskScape_Projects\`. Open a command prompt and `cd` to the directory where you unzipped the files, e.g. ```none cd C:\RiskScape_Projects\expressions-tutorial ``` You will use this command prompt to run the RiskScape commands in this tutorial. The unzipped project contains a few Python functions and a simple model. This data is similar to the :ref:`Upolu tsunami data ` that we used in the previous tutorials. .. note:: This input data was provided by `NIWA `_ and is based on published research papers. The data files have been adapted slightly for this tutorial. Most expressions we will use in this tutorial can be run *independently* of any project. However, this project will give us some practical examples to test out expressions. ## The basics All RiskScape models are based on an underlying data-processing *pipeline*. Pipelines are a series of linked data-processing steps. The processing in each step is typically described using RiskScape *expressions*. Expressions are simple statements - they resemble the formula you might enter into a spreadsheet cell, or python one-liners. Expressions allow you to _transform_ your input data as it flows through your model pipeline. This tutorial will use the `riskscape expression evaluate` command, which evaluates a single RiskScape expression and displays the result. This lets us test out RiskScape expressions before we use them in a pipeline. The command is similar to the Python interactive console, if you have used that before. .. tip:: You can also use the ``riskscape expr eval`` command alias to evaluate expressions. It behaves exactly the same as ``riskscape expression evaluate``, but is slightly shorter to type. Let's try it now with the simplest possible expression. Type the following into the command prompt where you would normally run RiskScape. ```none riskscape expression evaluate " 1 + 1 " ``` .. note:: The expression itself needs to be enclosed in double-quotes here, purely for the command-line. When you use an expression in a pipeline you will not need the double-quotes like this. We have left spaces around the double-quotes so that the standalone expression is easier to see, but make sure you copy both double-quotes. ## Types All the data used in a RiskScape model has an underlying :ref:`data type`. RiskScape expressions are the same - every part of the expression has a type. Common _simple_ types are `Text` strings, `Integer`, `Floating`, `Boolean`, and `Geometry`. RiskScape also supports _complex_ types, like `List` and `Struct`, which we will cover in more detail later. The type is important, as many operations will not work if one of the input types is incorrect. For example, adding two `Text` strings together is fine. ```none riskscape expression evaluate " 'abc' + 'xyz' " ``` Whereas adding together two _different_ types can produce an error. ```none riskscape expression evaluate " 1 + 'xyz' " Failed to evaluate expression - Operator '+' is not supported for types '[Integer, Text]' ``` Note let's try adding an `Integer` and a `Floating` value together. Remember that an `Integer` is always a whole number, whereas `Floating` values always have a decimal point. ```none riskscape expression evaluate " 1 + 2.5 " ``` In this case it works, even though the two numbers are different types. This is called _type coercion_ - RiskScape turned the `Integer` value into a `Floating` value. The rules of coercion are complicated, but it basically means that when two different types are compatible, RiskScape will 'make do' with the values it was given. .. tip:: You can see a list of all the built-in types that RiskScape supports using the ``riskscape type-registry list`` command. ## Function calls Of course, expressions can do much more interesting things than just adding two numbers together. In general, more complicated operations will involve a _function call_. If function calls are new to you, don't panic. They're a bit like the function formula you would use in a spreadsheet, e.g. `SUM()` or `MAX()`. A function call executes a named piece of code by giving it some specific values. The values you pass to the function are called the _function arguments_. RiskScape provides some common built-in functions for you, but you can also add whatever user-defined functions you want. For example, the following function call will generate a random number between zero and one. Try running it a few times and check you get different numbers. ```none riskscape expression evaluate " random_uniform(0, 1) " ``` .. tip:: Each built-in function comes with some basic documentation that will help you understand how to use it. You can read this help documentation using the ``riskscape function info FUNCTION_NAME`` command. You can see a list of supported functions using ``riskscape function list --all``. Try reading more about the `random_uniform()` function now, using the following command: ```none riskscape function info random_uniform ``` Did you notice that the function expects `Floating` argument-types, but we passed it `Integer` values? That's coercion at work again. Functions can have optional function arguments that can be omitted. You can also specify the argument _keyword_ names as part of the function call. As an example, `log()` accepts an optional 'base' argument, but defaults to natural log if no base is specified. ```none riskscape expression evaluate " log(10) " ``` Whereas specifying a different base changes the result (with the argument keywords used for clarity). ```none riskscape expression evaluate " log(x: 10, base: 10) " ``` .. tip:: You can call your own user-defined :ref:`functions` from any expression, just like you can with the built-in RiskScape functions. ## Conditional logic You may find cases where you want to use some basic if-then-else logic in your pipeline. RiskScape has an `if_then_else()` function to assist us in these cases. The `if_then_else()` function takes a `Boolean` (true or false) 'condition' argument, and then returns one of two values depending on whether or not the condition was true. As an example, here's an expression that simulates tossing a coin. Try running it a few times to check you get both heads and tails. ```none riskscape expr eval " if_then_else( random_uniform(0, 1) > 0.5, 'Heads', 'Tails' ) " ``` Here the _condition_ is whether a randomly generated number is greater than 0.5. If the condition is met, we say the coin toss was 'Heads'. If not, we say it was 'Tails'. .. note:: In this example we use the ``>`` greater-than operator. Other comparison operators you could use include ``>=``, ``<``, ``<=``, ``=`` (equal), and ``!=`` (not equal). As an aside, in this particular case we could use another function here instead to make the expression a lot simpler, e.g. ```none riskscape expression evaluate " random_choice( ['Heads', 'Tails'] ) " ``` .. tip:: If you are unsure of the exact function name, ``riskscape function info`` will suggest similar alternatives. For example, try entering ``riskscape function info random`` - it will list the functions with 'random' in their name in the error message. This can help you to discover what built-in functions RiskScape supports. ## Structs A *struct* is a RiskScape type that is used to group together a set of attributes. Attributes in your input data, such as shapefile data, are typically *simple* types, such as `Floating`, `Integer`, `Text`, etc. Whereas a `Struct` is a *compound* type that is made up of one or more other types, called struct _members_. You can define a new *instance* of a `Struct` type within a RiskScape expression, for example: ```none riskscape expression evaluate " { foo: 1 + 1, bar: 3.5 } " ``` This creates a struct instance with two *member* attributes: an `Integer` called `foo`, and a `Floating` called `bar`. .. tip:: You may notice that the syntax looks quite similar to defining a dictionary in Python. You can think of struct instances as being a lot like dictionaries in Python - both consist of a set of key-value pairs. In fact, when a RiskScape struct instance is passed to a Python function, it becomes a Python dictionary. Often RiskScape pipelines will use `as` to name the member attributes in the struct. The following expression produces the same struct as the previous example. ```none riskscape expression evaluate " { 1 + 1 as foo, 3.5 as bar } " ``` You can also use much more complicated expressions, such as function calls, when defining struct members, e.g. ```none riskscape expression evaluate " { foo: max(1, 2), bar: random_uniform(3.4, 3.6) } " ``` ### Using Structs You may often only be interested in one particular attribute, or struct member, rather than the whole struct. For example, trying to pass a whole struct to the `round()` function results in an error because the argument type does not match. ```none riskscape expression evaluate " round( { foo: 1, bar: 2.5 } ) " Failed to evaluate expression - Problems found with 'round' riskscape function - Type mismatch for 'number' function argument. Expected 'Floating' but found '{foo=>Integer, bar=>Floating}' ``` In this case, we can use the `.` operator to _access_ the struct member we're interested in (`bar`). E.g. ```none riskscape expression evaluate " round( { foo: 1, bar: 2.5 }.bar ) " ``` .. note:: This example looks a little strange because we're defining the struct and accessing a struct member at the same time. Normally accessing a struct member would look more like ``exposure.geom`` or ``consequence.loss``. Sometimes you can in fact pass the whole struct to a function, thanks once again to the magic of coercion. This only works if it's obvious from the `Struct` type what you want to do, such as when the struct only has one member, like the following example: ```none riskscape expression evaluate " round( { bar: 2.5 } ) " ``` ### Pipeline data These examples to create structs may not seem particularly exciting, but structs are a fundamental part of pipelines. Structs allow you to organize your pipeline data, so that you can transform the data more easily. For example, say your function needs values for 10 different attributes of a building before it can calculate its damage. It's a lot simpler to pass your function one 'building' struct as an argument, rather than 10 individual arguments. Every single row or record in your input data gets turned into a struct instance. These struct instances get _transformed_ (i.e. combined, split up, or manipulated) as they move from one pipeline step to the next. So each pipeline step produces a new struct instance as output, which then becomes the input struct instance for the next step. We call these struct instances _Tuples_, which represent a row of model input data. The following diagram shows a simple example where only three tuples, or rows of data, are shown. .. image:: ../diagrams/workflow_data.png :width: 80% :alt: Tuples shown before and after a Spatial Sampling pipeline step. Prior to the 'Spatial Sampling' pipeline step, the tuple contains an `exposure` struct that contains the exposure-layer input data. The type information for the tuple would look like: `{ exposure: { ID: integer, Use_Cat: text, Cons_Frame: text } }`. After the 'Spatial Sampling' pipeline step, a `hazard` attribute is added, that contains the hazard intensity measure sampled from the hazard-layer. The type information for the tuple would now look like: `{ exposure: { ID: integer, Use_Cat: text, Cons_Frame: text }, hazard: Floating }`. ## Casting For many simple types, you can change the underlying type using a _casting_ function, E.g. ```none riskscape expression evaluate " 1 + int('2') " riskscape expression evaluate " str(1) + 'xyz' " ``` You can also use `round()` to convert a `Floating` number to an `Integer` value, e.g. ```none riskscape expression evaluate " round(1.95) " ``` .. tip:: Input data from a CSV file will *always* be ``Text`` type. So you will need to use casts to turn any numeric CSV values into ``Integer`` or ``Floating`` values. This can be done in the bookmark INI definition, by using ``set-attribute.foo = int(bar)``. Refer to the :ref:`tutorial on bookmarks ` for examples of this. ## Lists Lists are declared as a comma-separated series of items enclosed in `[ ]` square brackets. This is similar syntax to Python, e.g. ```none riskscape expression evaluate " length( [1, 2, 3] ) " ``` ## Aggregation The `mean()` function is an :ref:`aggregation` operation - it takes a bunch of numbers and combines them into a single result (or a group of results). Other common aggregation functions include `max()`, `sum()`, `count()`, and `stddev()`. Aggregation expressions are a special type of expression. They mostly only get used in aggregation pipeline steps (i.e. the `group` step). Normally an expression only applies to one _row_ of input data at a time. Aggregation expressions are different in that they apply to a _column_ (or attribute) across _all_ rows of input data. Another way to think of it is using an aggregation expression in RiskScape is a lot like using `SUM()` on a spreadsheet column. Unfortunately, we can't test aggregating _rows_ of data using the `riskscape expression evaluate` command. However, we can still use the aggregation _functions_ on lists. Instead of aggregating _columns_ of input data, the function aggregates every item in the list. Try calling some aggregation functions with a simple list. ```none riskscape expression evaluate " mean( [1, 2, 3, 4, 5] ) " riskscape expression evaluate " median( [1, 2, 3, 4, 5] ) " riskscape expression evaluate " sum( [1, 2, 3, 4, 5] ) " riskscape expression evaluate " max( [1, 2, 3, 4, 5] ) " riskscape expression evaluate " percentile( [1, 2, 3, 4, 5], 90 ) " ``` .. note:: Aggregation can change the *type* of your data. You may have noticed that some aggregation functions always produce a `Floating` result, even though the list contains `Integer` values. This is because they use division, and dividing integers always results in a floating-point value. ## Nullable Occasionally you might notice that RiskScape says one of your types is `Nullable`. This happens when the underlying value may or may *not* exist. Usually a `Nullable` type occurs from sampling the hazard-layer with your exposure-layer geometry. The exposure geometry might fall completely outside the hazard-layer's bounds, in which case there is no possible value to return. So RiskScape returns `Nothing` (this is the equivalent of `None` in Python terms). The important thing to know is that `Nullable` types are _sticky_. Any other types that come into contact with a `Nullable` type get tainted by association. For example: ```none riskscape expression evaluate " { foo: 1 + null_of('integer') } " ``` In this case, the _type_ of `foo` is `Nullable` because part of the expression was `Nullable`. The _value_ of `foo` is actually null, or `Nothing`, because part of the expression was null. .. note:: RiskScape automatically checks your expressions for null safety. If a null value is used in an operation that does not expect null values then the result will be null. ## Assignment You may have noticed we have only mentioned the `=` equals sign once so far, and that was for checking equality, e.g. ```none riskscape expression evaluate " 1 = 1 " ``` RiskScape expressions never _update_ an existing value in a struct or tuple, and so you will never use `=` for assignment (setting a value). Instead, RiskScape always creates a _new_ struct instance with the modified value. This point may seem a bit academic (and it is - it's based on a [Functional Programming paradigm](https://en.wikipedia.org/wiki/Functional_programming#Referential_transparency)). All you need to know for now is that you will never use `=` to set values (nor will you use `+=` or `-=`). ## Practical applications We have covered the basic syntax of RiskScape expressions and how they work. Now let's look at some practical applications of using expressions in your models. ### Extending wizard models Many of the parameters that wizard models take are simply RiskScape expressions. The wizard will help you build *simple* expressions interactively, but these are fairly limited in what they can do. However, you can take a saved configuration for a wizard model and modify the expressions manually. Let's use the following model in the `project.ini` file as an example: ```ini [model example] description = Simple model with expressions that we can modify framework = wizard input-exposures.layer = data/Buildings_SE_Upolu.shp input-exposures.geoprocess = false input-hazards.layer = data/MaxEnv_All_Scenarios_50m.tif sample.hazards-by = CLOSEST analysis.function = is_exposed report-event-impact.filter = hazard > 0 report-event-impact.group-by[0] = exposure.Cons_Frame report-event-impact.aggregate[0] = count(exposure) as Total_buildings report-event-impact.aggregate[1] = mean(hazard) as Average_depth report-event-impact.select[0] = * ``` Currently the model reports the `Average_depth` in metres. If we wanted this result recorded in centimetres instead, the interactive wizard currently cannot help us build that expression. However, it would simply be `Average_depth * 100`. If we wanted to control the name that the new attribute gets, we could use `Average_depth * 100 as Average_depth_cm`. If we wanted to round the result to the nearest centimetre, we could use `round(Average_depth * 100)`. Try running the model using the following command: ```none riskscape model run example -p "report-event-impact.select[1]=round(Average_depth * 100) as Average_depth_cm" ``` This command adds an extra `Average_depth_cm` to the `event-impact.csv` results file, which should look like this: ```none Cons_Frame,Total_buildings,Average_depth,Average_depth_cm Reinforced Concrete,2,1.0038375854492188,100 Masonry,1608,2.3271826401130475,233 Steel,5,2.5543127775192263,255 Reinforced_Concrete,10,2.543464779853821,254 ,3,1.3111796379089355,131 Timber,431,3.2016472523284345,320 ``` .. tip:: Usually it's simpler to modify the expression in the ``project.ini`` file directly, rather than using the ``-p`` or ``--parameter`` CLI options to change the model parameter on the fly. ### Expression scope An expression in a model pipeline will operate on the *tuple*, i.e. the row of data currently being processed. This mean the expression can access any attributes that are currently present in the tuple. .. note:: As the tuple moves through the model pipeline, the attributes it contains will change. So the attributes that are available to your expression will depend on *where* in the pipeline the expression is used. The interactive wizard can help you determine what attributes you can use. Another alternative is to simply try to access an attribute that you know will not exist (e.g. `foo`), and RiskScape should tell you what attributes are available in the error message. For example, try running the following command: ```none riskscape model run example -p "report-event-impact.select[1]=foo" ``` This should produce the following error: ```none There was a problem with the parameters for wizard model - Failed to load the saved model. Some parameters specified may be invalid. If you have altered parameters manually, try going through the interactive wizard again - Problems found with 'report-event-impact.select' parameter - Failed to validate 'select({*, foo})' step for execution - Failed to validate expression '{*, foo}' against input type {Cons_Frame=>Text, Total_buildings=>Integer, Average_depth=>Nullable[Floating]} - Could not find 'foo' among [Cons_Frame, Total_buildings, Average_depth] ``` The last line lists the names of the available attributes, which are: `Cons_Frame, Total_buildings, Average_depth`. The `select` parameter is one of the last steps in the model, so these attributes correspond to what gets saved in the `event-impact.csv` file. Let's try the same thing with an earlier step in the model pipeline: ```none riskscape model run example -p "report-event-impact.filter=foo" ``` This time the available attributes are `exposure.the_geom, exposure.Use_Cat, exposure.Cons_Frame, exposure.area, exposure.perimeter, hazard, consequence`. .. tip:: Remember that ``exposure`` is a struct and the member attributes it contains correspond to the attributes in the exposure-layer input file. ### Aggregation expressions Let's take a look at using expressions to aggregate the results of the model. Open the `project.ini` file in Notepad, or your preferred text editor, and look at lines 10-12: ```ini report-event-impact.group-by[0] = exposure.Cons_Frame report-event-impact.aggregate[0] = count(exposure) as Total_buildings report-event-impact.aggregate[1] = mean(hazard) as Average_depth ``` The right-hand side of the `=` are all expressions that control how RiskScape will aggregate the model data. The `report-event-impact.group-by` model parameter specifies the attribute(s) that we group the `event-impact` results by, and the `report-event-impact.aggregate` parameter specifies how to aggregate the results. Try changing these three lines to the following: ```ini report-event-impact.group-by[0] = exposure.Use_Cat report-event-impact.aggregate[0] = sum(exposure.area) as Building_m2 report-event-impact.aggregate[1] = stddev(hazard) as Depth ``` Save the `project.ini` file, and re-run the model using the following command: ```none riskscape model run example ``` It should produce an `event-impact.csv` file that looks something like this: ```none Use_Cat,Building_m2,Depth.mean,Depth.stddev Outbuilding,24355.417724609375,2.331427180173883,1.768794405391399 Industrial,1392.920654296875,2.7654923796653748,0.8030687924796163 Commercial,3797.490966796875,3.719742706843785,1.7849776453776574 Commercial - not verified,400.925048828125,, Religious,17922.424560555002,1.8323415092059545,1.7242769315238138 Community - not verified,759.182861328125,5.0066187381744385,0.5218430592989689 Tourist Fale,2094.622802734375,3.8892281768477965,1.259292364541857 Commericial - not verified,184.406005859375,, Hotel,19877.2353515625,3.390313490472659,1.8243371156722499 Education,5523.36743164,3.395870625972748,1.7343286937078735 ,20.02490234375,1.3111796379089355,1.0634330871567976 Community,7328.067626953125,3.2658567169438237,1.298215930801334 Residential,149052.96362304688,2.2250018423603426,1.5794900709365927 Fale,6829.74169921875,2.7053862896051495,1.8508132278699894 ``` Notice that the results are now quite different: - The results are now grouped by the building `Use_Cat`, rather than `Cons_Frame`, due to the change in the `report-event-impact.group-by[0]` model parameter. - Instead of counting the exposed buildings, we now `sum` the building's area (`Building_m2`). - The `stddev()` aggregation function outputs both the mean (`Depth.mean`) *and* standard deviation (`Depth.stddev`). ### Testing functions So far we have used expressions to call *built-in* RiskScape functions. However, you can also use expressions to call your own, user-defined functions (i.e. ones that you add via the `project.ini` file). This can be a helpful way to test that your Python code is working correctly. The `project.ini` file contains a simple `exposure_level` function that can be used in RiskScape models. This function takes two arguments: an element-at-risk and the hazard intensity measure. Typically the element-at-risk will be some sort of `Struct` type, but here it can be `Anything`. We can simply pass the function `{ }` as an empty struct (i.e. a struct that has no member attributes). The hazard argument takes a `Nullable`, `Floating` type. Try calling the `exposure_level` function using the following command: ```none riskscape expression evaluate " exposure_level( { }, 1.5 ) " ``` Try a few different hazard values and check that the return value is sensible. When the exposure is not exposed to the hazard, the `hazard` argument will be `Nothing`, or null. You can test passing the function a null hazard value using the following statement: ```none riskscape expression evaluate " exposure_level( { }, null_of('floating') ) " ``` .. tip:: When you use CPython, you can also use the ``if __name__ == '__main__':`` Python idiom to run your Python code manually and test that it works correctly. Whereas using RiskScape expressions will work regardless of whether you are using CPython or Jython. #### Declaring struct types vs instances In the `project.ini` file, take a look at the `[function exposure_level]` section. Notice that how we define the `Struct` *type* for `return-type` looks quite different to how we have been creating struct instances so far. This difference is because the INI file defines an abstract 'blueprint' for the `Struct` type, whereas our expressions have created concrete instances of the struct. In the `project.ini` file we define what the `Struct` type looks like *without* any values for the member attributes. Whereas, with expressions we *only* have attribute values and RiskScape infers what type the member attributes must be. #### Testing with element-at-risk data Let's look at another function, `Samoa_Building_Fragility_DS5`, which requires that a specific attribute (`Cons_Frame`) is present for the element-at-risk. Its Python code looks like this: ```python def function(building, hazard_depth): DS_5_Prob = 0.0 construction = building.get("Cons_Frame") if hazard_depth is not None and hazard_depth > 0: if construction == 'Steel': DS_5_Prob = log_normal_cdf(hazard_depth, 0.39, 0.4) elif construction == 'Masonry': DS5_Prob = log_normal_cdf(hazard_depth, 0.39, 0.4) elif construction in ['Reinforced_Concrete', 'Reinforced Concrete']: DS_5_Prob = log_nomral_cdf(hazard_depth, 0.86, 0.94) else: # 'Timber' or unknown DS_5_Prob = log_normal_cdf(hazard_depth, 0.1, 0.28) return DS_5_Prob def log_normal_cdf(x, mean, stddev): # this uses the built-in RiskScape 'lognorm_cdf' function return functions.get('lognorm_cdf').call(x, mean, stddev) ``` Using what we have learnt so far, we can create a struct instance that has a `Cons_Frame` attribute and pass that to our function. Try running the following command: ```none riskscape expression evaluate " Samoa_Building_Fragility_DS5( { Cons_Frame: 'Timber' }, 1.1 ) " ``` Try varying the `Cons_Frame` value to check that the function behaves as expected. Note that: - The `Cons_Frame` can be `Timber`, `Steel`, `Masonry`, or `Reinforced Concrete`. - Timber, Masonry, and reinforced concrete should all produce different results for the same hazard value. - If `Cons_Frame` is unknown, then currently the function defaults to timber. .. note:: This function contains *two* fundamental bugs. See if you can find them by using the ``riskscape expression evaluate`` command to test the function. One bug would prevent the model from running properly, whereas the other bug will produce the wrong model results. ### Versatility of functions In the tutorials so far, the focus has been on using Python functions to analyse the consequence that the hazard has on each element-at-risk. However, you can add your own Python functions and use them from *any* RiskScape expression. Functions give you a lot of versatility as to what you can do in a model. For example, if you wanted to *pre-process* your input data or *post-process* your results further. In your `project.ini` bookmark definitions, the `set-attribute.ATTRIBUTE_NAME` parameter simply takes a RiskScape expression. So you could call your own user-defined Python function there to process the input data *before* it gets used in the model. This lets you 'clean' the input data as part of the model, or reclassify attribute values to handle schema differences in the input data. Similarly, in wizard models the `report-event-impact.select` parameter also takes any expression. So you could also call a Python function there to process the results of the consequence analysis further. .. tip:: If you want to use RiskScape functions in other places in your model besides the Consequence Analysis phase, there are :ref:`Hello, world example ` for constructing RiskScape functions with different argument types. ## Recap We have covered a lot, so here are some key takeaway points: - Everything in RiskScape has a type. - The actual data type needs to match what is expected. - RiskScape can _sometimes_ make do when two types are close enough (coercion). - Structs help to organize your data as it flows through the pipeline. - Each pipeline step _transforms_ your data in some way. - Function calls are the real work-horse when it comes to transforming your data. - Most expressions only transform _one_ row of data at a time. However, aggregation expressions work across _all_ rows of data. - The `Nullable` type is 'sticky' and can taint other data types by association. - You use `=` for checking for equality, not `==`. You never use `=` to _set_ a value. - Many parameters in a wizard model are simply RiskScape expressions. Manually editing the expressions in the saved model configuration can give you better flexibility over what your model does. - You can add your own Python functions and use them from *any* RiskScape expression. This gives you a lot of versatility as to what you can do in a RiskScape model, such as pre-processing or post-processing the data. - You can use the `riskscape expression evaluate` command to test that your function works correctly given various different arguments. If you want to learn more about writing RiskScape _pipelines_, you could now look at the :ref:`pipelines_tutorial` tutorial. ## Extra for experts If you want to explore using RiskScape expressions a little further, you could try the following exercises out on your own. - By changing the arguments, try using the `random_uniform()` function to generate a random floating-point number between 1 and 10. Try passing the result to the `round()` function to generate a random *integer* number between 1 and 10. _Tip_: the arguments for one function can be the return-value from another function. - The `map()` function is a handy way to transform or manipulate a given list of values. The second function argument is a *lambda* expression in the form `list_item -> new_value`. The lambda expression describes how the list item should be transformed. Try evaluating the expressions `map( [1, 2, 3], x -> x * 2 )` and `map( [1, 2, 3], x -> pow(x, 2) )` and see what they do. - Using the `map()` and `random_uniform()` functions, see if you can generate a list of 5 random numbers. How about 50 random numbers? _Hint_: RiskScape has a `range()` function that generates a list of numbers, similar to `range()` in Python. - In the `example` model, try changing the `sum(exposure.area)` expression so that the result is rounded. There are two ways to do this - rounding each individual building area *before* you sum it, i.e. `sum(round(exposure.area))`, or rounding the total building area *after* you have summed it, i.e. `round(sum(exposure.area))`. The approach you choose will make a difference to the result you get. - Try out the following aggregation concepts by modifying and running the `example` model. - `count(*)` is a simple way to count everything. - If you pass a boolean expression to `count()`, it works like the `COUNTIF` spreadsheet formula does, e.g. ```none count(hazard > 0.5) ``` - Common aggregation functions include `max`, `min`, `mean`, `median`, `mode`, `stddev`, and `sum`. Try out any functions that you are interested in. - There is also the `percentile` function that takes a second argument, which is the percentile to use, e.g. `percentile(hazard, 90)`. There is also `percentiles`, which takes a list of percentiles, e.g. `percentile(hazard, [10, 25, 75, 90])`. - The `bucket_range()` function is an advanced feature, but quite powerful. It lets you assign your data into [buckets, or bins](https://en.wikipedia.org/wiki/Data_binning), and aggregate based on the bucket it falls into. Try out the expression: ```none bucket_range(pick: hazard, select: { count(*) }, range: [ 1, 2, 3 ]) as Depth ``` This counts the number of buildings based on the range that the hazard value falls: <1m, 1-2m, 2-3m, 3m+.