From 8acec1d5beab2829e53677a96f981f1bcd05bafa Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Thu, 14 Aug 2025 10:28:55 +0200
Subject: [PATCH 01/13] added example with IF modeling

---
 tests/test_modeling.py | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/tests/test_modeling.py b/tests/test_modeling.py
index fb13230..fa89c8f 100644
--- a/tests/test_modeling.py
+++ b/tests/test_modeling.py
@@ -132,6 +132,29 @@ def test_total_person_hours(self):
         np.testing.assert_allclose(sample_mean, expected_mean, rtol=0.02)
         np.testing.assert_allclose(sample_std, expected_std, rtol=0.02)
 
+    def test_conditional_if_statement(self):
+        """Suppose mens height has distribution N(176, 7.1).
+        What is the distribution of the difference between height of two men?
+        Caveat: there is a 10% chance that the two men are identical twins,
+        and in that case their height should be perfectly equal.
+        """
+
+        # Height of two random men
+        height1 = Distribution("norm", loc=176, scale=7.1)
+        height2 = Distribution("norm", loc=176, scale=7.1)
+
+        # If they are twins, their height should be perfectly correlated
+        is_twin = Distribution("bernoulli", p=0.1)
+
+        # height2 = IF(is_twin, height1, height2)
+        height2 = is_twin * height1 + (1 - is_twin) * height2
+
+        # This is the answer to the question
+        (abs(height2 - height1)).sample(999, random_state=42)
+
+        # At least one of the realizations should be idential
+        assert np.any(np.isclose(height1.samples_, height2.samples_))
+
 
 def test_copying():
     # Create a graph

From 8d15c1d3a0f2fe63f6d481839fd9fba33fd186d9 Mon Sep 17 00:00:00 2001
From: Tommy Odland <10076072+tommyod@users.noreply.github.com>
Date: Thu, 14 Aug 2025 10:33:31 +0200
Subject: [PATCH 02/13] Update tests/test_modeling.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---
 tests/test_modeling.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/test_modeling.py b/tests/test_modeling.py
index fa89c8f..ef263e5 100644
--- a/tests/test_modeling.py
+++ b/tests/test_modeling.py
@@ -152,7 +152,7 @@ def test_conditional_if_statement(self):
         # This is the answer to the question
         (abs(height2 - height1)).sample(999, random_state=42)
 
-        # At least one of the realizations should be idential
+        # At least one of the realizations should be identical
         assert np.any(np.isclose(height1.samples_, height2.samples_))
 
 

From 2414994611eeed9e9c675b56d13aed149e3b0551 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:08:46 +0200
Subject: [PATCH 03/13] first draft without corr

---
 src/probabilit/config.yml                  | 20 +++++++++++
 src/probabilit/read_yaml_create_samples.py | 40 ++++++++++++++++++++++
 2 files changed, 60 insertions(+)
 create mode 100644 src/probabilit/config.yml
 create mode 100644 src/probabilit/read_yaml_create_samples.py

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
new file mode 100644
index 0000000..ee855aa
--- /dev/null
+++ b/src/probabilit/config.yml
@@ -0,0 +1,20 @@
+
+metadata:
+  name: "Comprehensive Project Risk Analysis"
+  description: "Complete risk analysis including direct costs, quality impacts, schedule effects, market variations, and risk adjustments"
+  samples: 9
+  sampling_method: "latin_hypercube"
+
+variables:
+  # Direct Material and Equipment
+  Steel_Cost:
+    type: "norm"
+    parameters:
+      loc: 800000
+      scale: 1200000
+
+  Maintenance_Cost:
+    type: "uniform"
+    parameters:
+      loc: 0
+      scale: 1
\ No newline at end of file
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
new file mode 100644
index 0000000..a13de2d
--- /dev/null
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -0,0 +1,40 @@
+from probabilit.modeling import Distribution, Add
+import yaml
+from yaml.loader import Loader
+import pandas as pd
+
+file = "config.yml"
+
+
+# Load data from file into a dictionary
+with open(file, "r") as file_handle:
+    yaml_data = yaml.load(file_handle, Loader)
+    
+# Load sections of the dictionary as variables
+metadata = yaml_data["metadata"]
+variables = yaml_data["variables"]
+
+# Convert dict of {name:data, ...} to {name:Distribution, ...}
+for variable_name in variables.keys():
+    variable_data = variables[variable_name]
+    variables[variable_name] = Distribution(variable_data["type"],
+                            **variable_data["parameters"])
+    
+    
+# Dummy expression to sample all parents
+expression = Add(*variables.values())
+expression.sample(size=metadata["samples"])
+
+# Collect all samples into a dataframe
+df_samples = pd.DataFrame({name:distr.samples_ for (name, distr) in variables.items()})
+
+print(df_samples)
+
+
+
+
+
+
+    
+    
+    
\ No newline at end of file

From 1138279115cd86fd2c4d40f3015c0e165f2405f7 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:21:42 +0200
Subject: [PATCH 04/13] added correlations

---
 src/probabilit/config.yml                  | 13 ++++-
 src/probabilit/read_yaml_create_samples.py | 60 +++++++++++++---------
 2 files changed, 48 insertions(+), 25 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index ee855aa..9f4578d 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -17,4 +17,15 @@ variables:
     type: "uniform"
     parameters:
       loc: 0
-      scale: 1
\ No newline at end of file
+      scale: 1
+      
+  Height:
+    type: "norm"
+    parameters:
+      loc: 180
+      scale: 10
+      
+      
+correlations:
+  - 0.2: [Steel_Cost, Maintenance_Cost]
+  - 0.5: [Height, Maintenance_Cost]
\ No newline at end of file
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index a13de2d..67269d5 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -2,39 +2,51 @@
 import yaml
 from yaml.loader import Loader
 import pandas as pd
+import numpy as np
+import scipy as sp
 
-file = "config.yml"
+if __name__ == "__main__":
+    file = "config.yml"
 
+    # Load data from file into a dictionary
+    with open(file, "r") as file_handle:
+        yaml_data = yaml.load(file_handle, Loader)
 
-# Load data from file into a dictionary
-with open(file, "r") as file_handle:
-    yaml_data = yaml.load(file_handle, Loader)
-    
-# Load sections of the dictionary as variables
-metadata = yaml_data["metadata"]
-variables = yaml_data["variables"]
+    # Load sections of the dictionary as variables
+    metadata = yaml_data["metadata"]
+    variables = yaml_data["variables"]
+    correlations = yaml_data.get("correlations", [])
 
-# Convert dict of {name:data, ...} to {name:Distribution, ...}
-for variable_name in variables.keys():
-    variable_data = variables[variable_name]
-    variables[variable_name] = Distribution(variable_data["type"],
-                            **variable_data["parameters"])
-    
-    
-# Dummy expression to sample all parents
-expression = Add(*variables.values())
-expression.sample(size=metadata["samples"])
+    # Convert dict of {name:data, ...} to {name:Distribution, ...}
+    for varname in variables.keys():
+        vardata = variables[varname]
+        variables[varname] = Distribution(vardata["type"], **vardata["parameters"])
 
-# Collect all samples into a dataframe
-df_samples = pd.DataFrame({name:distr.samples_ for (name, distr) in variables.items()})
+    # Dummy expression to sample all parents
+    expression = Add(*variables.values())
 
-print(df_samples)
+    # Add every correlation pair
+    # NOTE if we define correlations between (a, b) and (c, d)
+    # then the non-specified correlations (e.g. (a, c)) will be optimized
+    # towards zero.
+    for correlation in correlations:
+        value, (varname1, varname2) = next(iter(correlation.items()))
 
+        # Prepare input
+        corr_mat = sp.linalg.circulant([1.0, value])
+        var1, var2 = variables[varname1], variables[varname2]
 
+        # Add correlation pair
+        expression.correlate(var1, var2, corr_mat=corr_mat)
 
+    # Samle the expression, which populates `.samples_` on Distributions
+    expression.sample(size=metadata["samples"])
 
+    # Collect all samples into a dataframe
+    df_samples = pd.DataFrame(
+        {name: distr.samples_ for (name, distr) in variables.items()}
+    )
 
+    print(df_samples)
 
-    
-    
-    
\ No newline at end of file
+    print(df_samples.corr())

From f479281c40318b7f38f56dcba5483279aec767ad Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:31:26 +0200
Subject: [PATCH 05/13] non-scipy distrs

---
 src/probabilit/config.yml                  |  7 +++++++
 src/probabilit/read_yaml_create_samples.py | 24 +++++++++++++++++++---
 2 files changed, 28 insertions(+), 3 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index 9f4578d..34f7cec 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -25,6 +25,13 @@ variables:
       loc: 180
       scale: 10
       
+  Children:
+    type: "EmpiricalDistribution"
+    parameters:
+      data: [1, 3, 2, 4, 5, 2, 3, 4]
+      method: "closest_observation"
+  
+      
       
 correlations:
   - 0.2: [Steel_Cost, Maintenance_Cost]
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index 67269d5..7215e1f 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -1,10 +1,19 @@
-from probabilit.modeling import Distribution, Add
+from probabilit.modeling import (
+    Distribution,
+    Add,
+    EmpiricalDistribution,
+    CumulativeDistribution,
+)
 import yaml
 from yaml.loader import Loader
 import pandas as pd
-import numpy as np
 import scipy as sp
 
+TYPE_MAPPING = {
+    "empiricaldistribution": EmpiricalDistribution,
+    "cumulativedistribution": CumulativeDistribution,
+}
+
 if __name__ == "__main__":
     file = "config.yml"
 
@@ -20,7 +29,16 @@
     # Convert dict of {name:data, ...} to {name:Distribution, ...}
     for varname in variables.keys():
         vardata = variables[varname]
-        variables[varname] = Distribution(vardata["type"], **vardata["parameters"])
+        var_type = vardata["type"].lower()
+
+        # See if variable matches one of the non-scipy distributions first
+        if var_type in TYPE_MAPPING.keys():
+            var_class = TYPE_MAPPING[var_type]
+            variables[varname] = var_class(**vardata["parameters"])
+            continue
+
+        # Not a non-scipy distribution, so try scipy next
+        variables[varname] = Distribution(var_type, **vardata["parameters"])
 
     # Dummy expression to sample all parents
     expression = Add(*variables.values())

From 4ec53610a4b024cf93bea3ef5c1e9238152eb384 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:37:36 +0200
Subject: [PATCH 06/13] remove 'parameters', add seed and sampling method

---
 src/probabilit/config.yml                  | 25 ++++++++++------------
 src/probabilit/read_yaml_create_samples.py | 20 +++++++++++++----
 2 files changed, 27 insertions(+), 18 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index 34f7cec..fe659b4 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -2,34 +2,31 @@
 metadata:
   name: "Comprehensive Project Risk Analysis"
   description: "Complete risk analysis including direct costs, quality impacts, schedule effects, market variations, and risk adjustments"
-  samples: 9
-  sampling_method: "latin_hypercube"
+  samples: 999
+  sampling_method: "lhs"
+  seed: 42
 
 variables:
   # Direct Material and Equipment
   Steel_Cost:
     type: "norm"
-    parameters:
-      loc: 800000
-      scale: 1200000
+    loc: 800000
+    scale: 1200000
 
   Maintenance_Cost:
     type: "uniform"
-    parameters:
-      loc: 0
-      scale: 1
+    loc: 0
+    scale: 1
       
   Height:
     type: "norm"
-    parameters:
-      loc: 180
-      scale: 10
+    loc: 180
+    scale: 10
       
   Children:
     type: "EmpiricalDistribution"
-    parameters:
-      data: [1, 3, 2, 4, 5, 2, 3, 4]
-      method: "closest_observation"
+    data: [1, 3, 2, 4, 5, 2, 3, 4]
+    method: "closest_observation"
   
       
       
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index 7215e1f..78758f6 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -17,6 +17,8 @@
 if __name__ == "__main__":
     file = "config.yml"
 
+    # =================== LOAD ===================
+
     # Load data from file into a dictionary
     with open(file, "r") as file_handle:
         yaml_data = yaml.load(file_handle, Loader)
@@ -26,19 +28,23 @@
     variables = yaml_data["variables"]
     correlations = yaml_data.get("correlations", [])
 
+    # =================== CONVERT ===================
+
     # Convert dict of {name:data, ...} to {name:Distribution, ...}
     for varname in variables.keys():
         vardata = variables[varname]
-        var_type = vardata["type"].lower()
+        var_type = vardata.pop("type").lower()
 
         # See if variable matches one of the non-scipy distributions first
         if var_type in TYPE_MAPPING.keys():
             var_class = TYPE_MAPPING[var_type]
-            variables[varname] = var_class(**vardata["parameters"])
+            variables[varname] = var_class(**vardata)
             continue
 
         # Not a non-scipy distribution, so try scipy next
-        variables[varname] = Distribution(var_type, **vardata["parameters"])
+        variables[varname] = Distribution(var_type, **vardata)
+
+    # =================== CORRELATIONS ===================
 
     # Dummy expression to sample all parents
     expression = Add(*variables.values())
@@ -57,8 +63,14 @@
         # Add correlation pair
         expression.correlate(var1, var2, corr_mat=corr_mat)
 
+    # =================== SAMPLE ===================
+
     # Samle the expression, which populates `.samples_` on Distributions
-    expression.sample(size=metadata["samples"])
+    expression.sample(
+        size=metadata["samples"],
+        random_state=metadata["seed"],
+        method=metadata["sampling_method"],
+    )
 
     # Collect all samples into a dataframe
     df_samples = pd.DataFrame(

From a67b1ddf1f5d24bb2e87f10d9722a8b306227288 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:52:38 +0200
Subject: [PATCH 07/13] plotting :)

---
 src/probabilit/config.yml                  |  9 ++++++++-
 src/probabilit/read_yaml_create_samples.py | 12 +++++++++++-
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index fe659b4..60fcfec 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -4,6 +4,7 @@ metadata:
   description: "Complete risk analysis including direct costs, quality impacts, schedule effects, market variations, and risk adjustments"
   samples: 999
   sampling_method: "lhs"
+  # correlation_method: ""
   seed: 42
 
 variables:
@@ -32,4 +33,10 @@ variables:
       
 correlations:
   - 0.2: [Steel_Cost, Maintenance_Cost]
-  - 0.5: [Height, Maintenance_Cost]
\ No newline at end of file
+  - 0.5: [Height, Maintenance_Cost]
+  
+  
+plot:
+  - [Steel_Cost, Maintenance_Cost]
+  - Steel_Cost
+  - [Steel_Cost, Maintenance_Cost, Height, Children]
\ No newline at end of file
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index 78758f6..9ffcc97 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -4,6 +4,7 @@
     EmpiricalDistribution,
     CumulativeDistribution,
 )
+import seaborn
 import yaml
 from yaml.loader import Loader
 import pandas as pd
@@ -27,6 +28,7 @@
     metadata = yaml_data["metadata"]
     variables = yaml_data["variables"]
     correlations = yaml_data.get("correlations", [])
+    plots = yaml_data.get("plot", [])
 
     # =================== CONVERT ===================
 
@@ -78,5 +80,13 @@
     )
 
     print(df_samples)
-
     print(df_samples.corr())
+    
+    # =================== PLOTS ===================
+    
+    for vars_plot in plots:
+        vars_plot = [vars_plot] if isinstance(vars_plot, str) else vars_plot
+        df = pd.DataFrame(
+            {var_plot: variables[var_plot].samples_ for var_plot in vars_plot})
+        seaborn.pairplot(df)
+        

From 124eed822058b9a3638da6387c36daf740b76fe3 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:55:07 +0200
Subject: [PATCH 08/13] prettier config

---
 src/probabilit/config.yml | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index 60fcfec..23b79c7 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -1,14 +1,12 @@
-
 metadata:
-  name: "Comprehensive Project Risk Analysis"
-  description: "Complete risk analysis including direct costs, quality impacts, schedule effects, market variations, and risk adjustments"
+  name: "Name for a test"
+  description: "Description for a test"
   samples: 999
   sampling_method: "lhs"
   # correlation_method: ""
   seed: 42
 
 variables:
-  # Direct Material and Equipment
   Steel_Cost:
     type: "norm"
     loc: 800000
@@ -28,14 +26,11 @@ variables:
     type: "EmpiricalDistribution"
     data: [1, 3, 2, 4, 5, 2, 3, 4]
     method: "closest_observation"
-  
-      
-      
+
 correlations:
   - 0.2: [Steel_Cost, Maintenance_Cost]
   - 0.5: [Height, Maintenance_Cost]
-  
-  
+
 plot:
   - [Steel_Cost, Maintenance_Cost]
   - Steel_Cost

From 99b3bef0f2953cffd74a0c9b2e39809f3794b9ef Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Tue, 19 Aug 2025 13:57:41 +0200
Subject: [PATCH 09/13] ruff

---
 src/probabilit/read_yaml_create_samples.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index 9ffcc97..b6e062f 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -81,12 +81,12 @@
 
     print(df_samples)
     print(df_samples.corr())
-    
+
     # =================== PLOTS ===================
-    
+
     for vars_plot in plots:
         vars_plot = [vars_plot] if isinstance(vars_plot, str) else vars_plot
         df = pd.DataFrame(
-            {var_plot: variables[var_plot].samples_ for var_plot in vars_plot})
+            {var_plot: variables[var_plot].samples_ for var_plot in vars_plot}
+        )
         seaborn.pairplot(df)
-        

From 6e68da4f82d47d540e750fb2c5b0809d30071360 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Wed, 20 Aug 2025 10:53:26 +0200
Subject: [PATCH 10/13] add derived variables

---
 src/probabilit/config.yml                  | 12 ++++++++++++
 src/probabilit/modeling.py                 |  7 +++++++
 src/probabilit/read_yaml_create_samples.py | 18 ++++++++++++++++--
 3 files changed, 35 insertions(+), 2 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index 23b79c7..984be47 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -26,10 +26,22 @@ variables:
     type: "EmpiricalDistribution"
     data: [1, 3, 2, 4, 5, 2, 3, 4]
     method: "closest_observation"
+    
+  Adult:
+    type: "DiscreteDistribution"
+    values: ["no", "yes"]
+    probabilities: [0.4, 0.6]
 
 correlations:
   - 0.2: [Steel_Cost, Maintenance_Cost]
   - 0.5: [Height, Maintenance_Cost]
+  
+derived:
+  Adult: # Derived FROM this variable
+    TypicalAge: # The derived parameter
+      - "no": 16 # mapping: From -> To
+      - "yes": 32
+  
 
 plot:
   - [Steel_Cost, Maintenance_Cost]
diff --git a/src/probabilit/modeling.py b/src/probabilit/modeling.py
index 6a946f5..576f94d 100644
--- a/src/probabilit/modeling.py
+++ b/src/probabilit/modeling.py
@@ -828,6 +828,13 @@ def _sample(self, size=None):
         return np.average(np.vstack(samples), axis=0)
 
 
+class NoOp(VariadicTransform):
+    """Sample all ancestor variables, but do nothing else."""
+
+    def _sample(self, size=None):
+        tuple(parent.samples_ for parent in self.parents)
+
+
 class BinaryTransform(Transform):
     """Class for binary transforms, such as Divide, Power, Subtract, etc."""
 
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index b6e062f..36e4c55 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -3,16 +3,21 @@
     Add,
     EmpiricalDistribution,
     CumulativeDistribution,
+    DiscreteDistribution,
+    NoOp,
 )
 import seaborn
 import yaml
 from yaml.loader import Loader
 import pandas as pd
 import scipy as sp
+import operator
+import functools
 
 TYPE_MAPPING = {
     "empiricaldistribution": EmpiricalDistribution,
     "cumulativedistribution": CumulativeDistribution,
+    "discretedistribution": DiscreteDistribution,
 }
 
 if __name__ == "__main__":
@@ -29,6 +34,7 @@
     variables = yaml_data["variables"]
     correlations = yaml_data.get("correlations", [])
     plots = yaml_data.get("plot", [])
+    derived = yaml_data.get("derived", [])
 
     # =================== CONVERT ===================
 
@@ -49,7 +55,7 @@
     # =================== CORRELATIONS ===================
 
     # Dummy expression to sample all parents
-    expression = Add(*variables.values())
+    expression = NoOp(*variables.values())
 
     # Add every correlation pair
     # NOTE if we define correlations between (a, b) and (c, d)
@@ -79,8 +85,16 @@
         {name: distr.samples_ for (name, distr) in variables.items()}
     )
 
+    # =================== DERIVED ===================
+    for derived_from, data in derived.items():
+        for derived_to in data.keys():
+            mapping = functools.reduce(operator.ior, data[derived_to])
+            df_samples = df_samples.assign(
+                **{derived_to: lambda df: df[derived_from].map(mapping)}
+            )
+
     print(df_samples)
-    print(df_samples.corr())
+    print(df_samples.select_dtypes("number").corr())
 
     # =================== PLOTS ===================
 

From 0230d4017d558358f03cdb9842f26b7fc164a7c9 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Wed, 20 Aug 2025 10:56:34 +0200
Subject: [PATCH 11/13] cleanup a bit

---
 src/probabilit/config.yml                  | 2 --
 src/probabilit/read_yaml_create_samples.py | 4 +++-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/probabilit/config.yml b/src/probabilit/config.yml
index 984be47..f421bb7 100644
--- a/src/probabilit/config.yml
+++ b/src/probabilit/config.yml
@@ -41,8 +41,6 @@ derived:
     TypicalAge: # The derived parameter
       - "no": 16 # mapping: From -> To
       - "yes": 32
-  
-
 plot:
   - [Steel_Cost, Maintenance_Cost]
   - Steel_Cost
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index 36e4c55..e066332 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -1,6 +1,5 @@
 from probabilit.modeling import (
     Distribution,
-    Add,
     EmpiricalDistribution,
     CumulativeDistribution,
     DiscreteDistribution,
@@ -86,8 +85,11 @@
     )
 
     # =================== DERIVED ===================
+    # Derived variables AFTER correlations, since inducing correlations
+    # on derived variables would break the connection.
     for derived_from, data in derived.items():
         for derived_to in data.keys():
+            # Create mapping {from1: to1, from2: to2, ...} and apply it
             mapping = functools.reduce(operator.ior, data[derived_to])
             df_samples = df_samples.assign(
                 **{derived_to: lambda df: df[derived_from].map(mapping)}

From e1302c02c325d3ceb5d142a53e8a40d0cf384202 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Fri, 22 Aug 2025 12:48:00 +0200
Subject: [PATCH 12/13] command line tool

---
 src/probabilit/config_onebyone.yml         |   4 +
 src/probabilit/read_yaml_create_samples.py | 131 ++++++++++++++++++---
 2 files changed, 121 insertions(+), 14 deletions(-)
 create mode 100644 src/probabilit/config_onebyone.yml

diff --git a/src/probabilit/config_onebyone.yml b/src/probabilit/config_onebyone.yml
new file mode 100644
index 0000000..fdb39ee
--- /dev/null
+++ b/src/probabilit/config_onebyone.yml
@@ -0,0 +1,4 @@
+defaults:
+  Steel_Cost: 800000
+  Maintenance_Cost: 0.5
+  Adult: "yes"
\ No newline at end of file
diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index e066332..d5fabbb 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -12,6 +12,7 @@
 import scipy as sp
 import operator
 import functools
+import argparse
 
 TYPE_MAPPING = {
     "empiricaldistribution": EmpiricalDistribution,
@@ -19,21 +20,16 @@
     "discretedistribution": DiscreteDistribution,
 }
 
-if __name__ == "__main__":
-    file = "config.yml"
-
-    # =================== LOAD ===================
 
-    # Load data from file into a dictionary
-    with open(file, "r") as file_handle:
-        yaml_data = yaml.load(file_handle, Loader)
+def design_matrix(config, verbose=0):
+    """Given a dictionary config, returns a dataframe with samples."""
 
     # Load sections of the dictionary as variables
-    metadata = yaml_data["metadata"]
-    variables = yaml_data["variables"]
-    correlations = yaml_data.get("correlations", [])
-    plots = yaml_data.get("plot", [])
-    derived = yaml_data.get("derived", [])
+    metadata = config["metadata"]
+    variables = config["variables"]
+    correlations = config.get("correlations", [])
+    plots = config.get("plot", [])
+    derived = config.get("derived", [])
 
     # =================== CONVERT ===================
 
@@ -95,8 +91,11 @@
                 **{derived_to: lambda df: df[derived_from].map(mapping)}
             )
 
-    print(df_samples)
-    print(df_samples.select_dtypes("number").corr())
+    if verbose > 0:
+        print("========== DESIGN MATRIX ==========")
+        print(df_samples)
+        print("========== CORRELATIONS ==========")
+        print(df_samples.select_dtypes("number").corr())
 
     # =================== PLOTS ===================
 
@@ -106,3 +105,107 @@
             {var_plot: variables[var_plot].samples_ for var_plot in vars_plot}
         )
         seaborn.pairplot(df)
+
+    return df_samples
+
+
+def cmd_designmatr(args):
+    """Execute the subcommand."""
+    print(args)
+
+    # Load data from file into a dictionary
+    with open(args.config, "r") as file_handle:
+        config = yaml.load(file_handle, Loader)
+        print(f"Loaded: {args.config}")
+
+    df_samples = design_matrix(config, verbose=args.verbose)
+    df_samples.to_csv(args.output, index=False)
+    print(f"Saved: {args.output}")
+
+
+def one_by_one(df, defaults=None, verbose=0):
+    """Transform a (n, p) dataframe to (n x p, p), keeping
+    all but one variable (column) constant at a time."""
+    if defaults is None:
+        defaults = dict()
+
+    for key in defaults.keys():
+        if key not in df.columns:
+            raise ValueError("Default {key=} not in {df.columns=}")
+
+    # Average for numeric, mode for rest
+    averages = df.select_dtypes("number").mean().to_dict()
+    modes = df.select_dtypes("number").mode().T.to_dict()[0]
+    constants = (averages | modes) | defaults
+
+    assert set(constants.keys()) == set(df.columns)
+
+    if verbose:
+        print(f"Constants: {constants}")
+
+    dfs = []
+    for column in df.columns:
+        # Set all columns except the current one to a constant
+        avg_map = {k: v for (k, v) in constants.items() if k != column}
+        dfs.append(df.assign(**avg_map))
+
+    return pd.concat(dfs)
+
+
+def cmd_onebyone(args):
+    """Execute the subcommand."""
+    print(args)
+
+    df = pd.read_csv(args.designmatrix)
+    print(f"Loaded: {args.designmatrix}")
+
+    if args.config is not None:
+        with open(args.config, "r") as file_handle:
+            config = yaml.load(file_handle, Loader)
+            print(f"Loaded: {args.config}")
+            defaults = config["defaults"]  # Default map col -> const
+    else:
+        defaults = None
+
+    df = one_by_one(df, defaults=defaults, verbose=args.verbose)
+    df.to_csv(args.output, index=False)
+    print(f"Saved: {args.output}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        prog="ProgramName",
+        description="What the program does",
+        epilog="Text at the bottom of help",
+    )
+
+    subparsers = parser.add_subparsers(help="")
+
+    # Parse the 'designmatrix' subcommand
+    p1 = subparsers.add_parser(
+        "designmatrix", help="Creates a design matrix (random samples)."
+    )
+    p1.add_argument("config", help="A .yml config file.")
+    p1.add_argument("--output", help="An output .csv file.", default="designmatrix.csv")
+    p1.add_argument("--verbose", "-v", action="count", default=0)
+    p1.set_defaults(func=cmd_designmatr)
+
+    # Parse the 'onebyone' subcommand
+    p2 = subparsers.add_parser(
+        "onebyone", help="Transform a design matrix to a one-by-one matrix."
+    )
+    p2.add_argument("designmatrix", help="Input .csv file.", default="designmatrix.csv")
+    p2.add_argument("--config", help="A .yml config file.", default=None)
+    p2.add_argument(
+        "--output",
+        action="store",
+        type=str,
+        help="Output file name.",
+        default="onebyone.csv",
+    )
+    p2.add_argument("--verbose", "-v", action="count", default=0)
+    p2.set_defaults(func=cmd_onebyone)
+
+    # Parse args and pass to function
+    args = parser.parse_args()
+    args.func(args)

From 69d697063c507bfeb1e6977c5a337d4372aae167 Mon Sep 17 00:00:00 2001
From: Tommy Odland <todl@equinor.com>
Date: Fri, 22 Aug 2025 12:53:36 +0200
Subject: [PATCH 13/13] reorder

---
 src/probabilit/read_yaml_create_samples.py | 28 ++++++++++------------
 1 file changed, 13 insertions(+), 15 deletions(-)

diff --git a/src/probabilit/read_yaml_create_samples.py b/src/probabilit/read_yaml_create_samples.py
index d5fabbb..2f5cec6 100644
--- a/src/probabilit/read_yaml_create_samples.py
+++ b/src/probabilit/read_yaml_create_samples.py
@@ -109,20 +109,6 @@ def design_matrix(config, verbose=0):
     return df_samples
 
 
-def cmd_designmatr(args):
-    """Execute the subcommand."""
-    print(args)
-
-    # Load data from file into a dictionary
-    with open(args.config, "r") as file_handle:
-        config = yaml.load(file_handle, Loader)
-        print(f"Loaded: {args.config}")
-
-    df_samples = design_matrix(config, verbose=args.verbose)
-    df_samples.to_csv(args.output, index=False)
-    print(f"Saved: {args.output}")
-
-
 def one_by_one(df, defaults=None, verbose=0):
     """Transform a (n, p) dataframe to (n x p, p), keeping
     all but one variable (column) constant at a time."""
@@ -154,7 +140,6 @@ def one_by_one(df, defaults=None, verbose=0):
 
 def cmd_onebyone(args):
     """Execute the subcommand."""
-    print(args)
 
     df = pd.read_csv(args.designmatrix)
     print(f"Loaded: {args.designmatrix}")
@@ -172,6 +157,19 @@ def cmd_onebyone(args):
     print(f"Saved: {args.output}")
 
 
+def cmd_designmatr(args):
+    """Execute the subcommand."""
+
+    # Load data from file into a dictionary
+    with open(args.config, "r") as file_handle:
+        config = yaml.load(file_handle, Loader)
+        print(f"Loaded: {args.config}")
+
+    df = design_matrix(config, verbose=args.verbose)
+    df.to_csv(args.output, index=False)
+    print(f"Saved: {args.output}")
+
+
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(
         prog="ProgramName",