From 6abc5d2075c9f31f80c54fc5b2f6ce4164414747 Mon Sep 17 00:00:00 2001
From: Aleksandra Apolinarska <aapolina@ethz.ch>
Date: Wed, 5 Mar 2025 15:47:45 +0100
Subject: [PATCH 1/2] #192 first draft: parallel coordinates plot
---
src/aixd/visualisation/plotter.py | 41 +++++++++++++++++++++++++++++++
1 file changed, 41 insertions(+)
diff --git a/src/aixd/visualisation/plotter.py b/src/aixd/visualisation/plotter.py
index 89c30535..c338d791 100644
--- a/src/aixd/visualisation/plotter.py
+++ b/src/aixd/visualisation/plotter.py
@@ -1336,6 +1336,47 @@ class Plotter:
return self._output(fig, output_name=output_name or f"PerformanceSummary_{block.name}")
+ def plot_parallel_coordinates(
+ self,
+ data: pd.DataFrame,
+ value_ranges: dict[str, Tuple[float, float]],
+ color_by: Optional[str] = None,
+ output_name: Optional[str] = None,
+ ):
+ """
+ Parallel coordinates plot for the given data.
+
+ Parameters
+ ----------
+ data: [:class:`pd.DataFrame`]
+ Dataframe containing data to plot. The order of the parallel coordinates corresponds to the order of columns.
+ value_ranges: dict[str, Tuple[float, float]]
+ Dictionary with column names as keys and [min,max] as values.
+ color_by: str, optional
+ Name of the variable to color the lines by. If not specified, default color (grey) is used for all samples.
+ output_name : str, optional, default=None
+ Name of the output file. If None, the name is automatically generated from the data block name.
+
+ Returns
+ -------
+ Optional[:class:`plotly.graph_objects.Figure`]
+ Plotly figure object, if self.output is None, otherwise None.
+ """
+ # TODO: accommodate categorical variables
+ # TODO: add transformed flag
+ # TODO: add flag to make value_ranges optional, and derive defaults from dataset/datamodule, or from given data
+ # TODO: separate function for dataset and separate for generated samples?
+ # TODO: add "variables" argument so that user can supply a whole df and specify which columns to plot?
+
+ if color_by is not None:
+ cmin, cmax = value_ranges[color_by]
+ linesettings = dict(color=data[color_by], colorscale=px.colors.sequential.Turbo, cmin=cmin, cmax=cmax)
+ else:
+ linesettings = dict(color="grey")
+
+ fig = go.Figure(data=go.Parcoords(line=linesettings, dimensions=list([dict(range=value_ranges[col], label=col, values=data[col].tolist()) for col in data.columns])))
+ return self._output(fig, output_name=output_name or "ParallelCoordinates")
+
@staticmethod
def _open_fig(size: Tuple[int, int] = (1, 1), **kwargs) -> go.Figure:
"""Helper method to open a figure with the desired number of rows and columns."""
--
GitLab
From 09377bdf534914e2e4e4bc37c083e80075143157 Mon Sep 17 00:00:00 2001
From: Aleksandra Apolinarska <aapolina@ethz.ch>
Date: Thu, 6 Mar 2025 16:49:58 +0100
Subject: [PATCH 2/2] #192 added handling of multidim and categorical/bool
variables
---
src/aixd/data/data_objects.py | 1 +
src/aixd/visualisation/plotter.py | 98 +++++++++++++++++++++++++++----
2 files changed, 88 insertions(+), 11 deletions(-)
diff --git a/src/aixd/data/data_objects.py b/src/aixd/data/data_objects.py
index 48083301..d7538793 100644
--- a/src/aixd/data/data_objects.py
+++ b/src/aixd/data/data_objects.py
@@ -1012,6 +1012,7 @@ class DataBool(DataCategorical):
def __init__(self, name: str, **kwargs):
domain = kwargs.pop("domain", dd.Options(["True", "False"], type="categorical"))
super().__init__(name=name, domain=domain, **kwargs)
+ self.type = "bool"
class DataOrdinal(DataDiscrete):
diff --git a/src/aixd/visualisation/plotter.py b/src/aixd/visualisation/plotter.py
index c338d791..420480c6 100644
--- a/src/aixd/visualisation/plotter.py
+++ b/src/aixd/visualisation/plotter.py
@@ -1336,10 +1336,11 @@ class Plotter:
return self._output(fig, output_name=output_name or f"PerformanceSummary_{block.name}")
- def plot_parallel_coordinates(
+ def parallel_coordinates(
self,
data: pd.DataFrame,
- value_ranges: dict[str, Tuple[float, float]],
+ variables: Optional[List[str]] = None,
+ value_ranges: Optional[dict[str, Tuple[float, float]]] = None,
color_by: Optional[str] = None,
output_name: Optional[str] = None,
):
@@ -1349,9 +1350,14 @@ class Plotter:
Parameters
----------
data: [:class:`pd.DataFrame`]
- Dataframe containing data to plot. The order of the parallel coordinates corresponds to the order of columns.
- value_ranges: dict[str, Tuple[float, float]]
- Dictionary with column names as keys and [min,max] as values.
+ Dataframe containing data to plot.
+ This could be data from the dataset or new samples generated by the model.
+ The order of the parallel coordinates corresponds to the order of columns.
+ variables: List[str], optional
+ List of variable names to plot (a selection from the data given). If not specified, all columns are plotted.
+ value_ranges: dict[str, Tuple[float, float]], optional
+ Dictionary with column names (or variable names, if `variables` are given) as keys and [min,max] as values.
+ If not specified, the ranges are derived from the dataset (if `variables` are given), or from the provided data (if otherwise).
color_by: str, optional
Name of the variable to color the lines by. If not specified, default color (grey) is used for all samples.
output_name : str, optional, default=None
@@ -1362,19 +1368,89 @@ class Plotter:
Optional[:class:`plotly.graph_objects.Figure`]
Plotly figure object, if self.output is None, otherwise None.
"""
- # TODO: accommodate categorical variables
# TODO: add transformed flag
- # TODO: add flag to make value_ranges optional, and derive defaults from dataset/datamodule, or from given data
# TODO: separate function for dataset and separate for generated samples?
- # TODO: add "variables" argument so that user can supply a whole df and specify which columns to plot?
+
+ # if data is real or int --> calc value ranges
+ # if data is bool or cat --> set tickvals and ticktext
+ # need to create a dummy variable:
+ # https://stackoverflow.com/questions/64139316/plotly-how-to-insert-a-categorical-variable-into-a-parallel-coordinates-plot
+
+ df = data
+
+ all_parcoorddimensions = []
+ linesettings = None
+
+ if variables:
+ """
+ If variable names are given, some settings are derived from the data objects via Dataset.
+ """
+ all_dobjs = self.dataset.get_data_objects_by_name(variables)
+ all_column_names = list(chain(*[dobj.columns_df for dobj in all_dobjs]))
+ for dobj in all_dobjs:
+ if color_by is not None and dobj.name == color_by:
+ if dobj.dim > 1:
+ raise ValueError(f"Cannot color by multi-dimensional attribute {dobj.name}.")
+
+ for colname in dobj.columns_df:
+ if dobj.type in ["real", "integer"]:
+ if value_ranges is not None:
+ vrange = value_ranges[colname]
+ else:
+ vrange = [dobj.domain.min_value, dobj.domain.max_value]
+ parcoorddimension = dict(range=vrange, label=colname, values=df[colname].tolist())
+ if color_by is not None and dobj.name == color_by:
+ linesettings = dict(color=df[color_by], colorscale=color_divergent_centered, cmin=vrange[0], cmax=vrange[1])
+ elif dobj.type in ["bool", "categorical"]:
+ options = [str(val) for val in dobj.domain.array]
+ options.sort()
+ dummyindex = {opt: idx for idx, opt in enumerate(options)}
+ parcoorddimension = dict(
+ tickvals=list(dummyindex.values()), ticktext=list(dummyindex.keys()), label=colname, values=[dummyindex[str(val)] for val in df[colname].tolist()]
+ )
+ else:
+ raise ValueError(f"DataObject type {dobj.type} not supported for parallel coordinates plot.")
+ all_parcoorddimensions.append(parcoorddimension)
+
+ else:
+ """
+ If variable names are not given, some settings are derived from the dataframe directly.
+ """
+ all_column_names = df.columns
+ df = df.infer_objects()
+ for colname in all_column_names:
+ if df[colname].dtype in [
+ "int64",
+ "float64",
+ "float32",
+ "int32",
+ ]:
+ if value_ranges is not None:
+ vrange = value_ranges[colname]
+ else:
+ vrange = [df[colname].min(), df[colname].max()]
+ parcoorddimension = dict(range=vrange, label=colname, values=df[colname].tolist())
+ if color_by is not None and colname == color_by:
+ linesettings = dict(color=df[color_by], colorscale=color_divergent_centered, cmin=vrange[0], cmax=vrange[1])
+ elif df[colname].dtype in ["str", "object", "bool"]:
+ options = [str(val) for val in df[colname].unique()]
+ options.sort()
+ dummyindex = {opt: idx for idx, opt in enumerate(options)}
+ parcoorddimension = dict(
+ tickvals=list(dummyindex.values()), ticktext=list(dummyindex.keys()), label=colname, values=[dummyindex[str(val)] for val in df[colname].tolist()]
+ )
+ else:
+ raise ValueError(f"Data type {df[colname].dtype} not supported for parallel coordinates plot.")
+ all_parcoorddimensions.append(parcoorddimension)
if color_by is not None:
- cmin, cmax = value_ranges[color_by]
- linesettings = dict(color=data[color_by], colorscale=px.colors.sequential.Turbo, cmin=cmin, cmax=cmax)
+ if linesettings is None:
+ # linesettings not set yet --> it is a categorical/bool variable
+ linesettings = dict(color=df[color_by], colorscale=color_qualitative10)
else:
linesettings = dict(color="grey")
- fig = go.Figure(data=go.Parcoords(line=linesettings, dimensions=list([dict(range=value_ranges[col], label=col, values=data[col].tolist()) for col in data.columns])))
+ fig = go.Figure(data=go.Parcoords(line=linesettings, dimensions=all_parcoorddimensions))
return self._output(fig, output_name=output_name or "ParallelCoordinates")
@staticmethod
--
GitLab