Visualise pipelines¶
Kedro-Viz displays data and machine-learning pipelines in an informative way, emphasising the connections between datasets and nodes. It shows the structure of your Kedro pipeline. This exercise assumes that you have been following the Spaceflights tutorial.
Install Kedro-Viz¶
If you did not already install Kedro-Viz when you installed the tutorial project dependencies, then you can do so now by running:
pip install kedro-viz
Visualise a whole pipeline¶
You should be in your project root directory. Once Kedro-Viz is installed, to visualise your pipeline, run:
kedro viz
This command will run a server on http://127.0.0.1:4141 that will open up your visualisation on a browser. You should be able to see the following:
You can also use the --autoreload
flag to autoreload Kedro Viz when a Python
or YAML
file has changed in the corresponding Kedro project.
If a visualisation panel opens up and a pipeline is not visible, then please check that your pipeline definition is complete. All other errors can be logged as GitHub Issues on the Kedro-Viz repository.
Exit an open visualisation¶
To exit this visualisation, close the open browser and enter Ctrl+C or Cmd+C in your terminal.
Visualise layers¶
A pipeline can be broken up into different layers according to how data is processed, and using a convention for layers makes it easier to collaborate. For example, the data engineering convention shown here labels datasets according to the stage of the pipeline (e.g. whether the data has been cleaned).
Kedro-Viz makes it easy to visualise these data processing stages by adding a layer
attribute to the datasets in the Data Catalog. We will be modifying the catalog.yml
file with the following:
companies:
type: pandas.CSVDataSet
filepath: data/01_raw/companies.csv
layer: raw
reviews:
type: pandas.CSVDataSet
filepath: data/01_raw/reviews.csv
layer: raw
shuttles:
type: pandas.ExcelDataSet
filepath: data/01_raw/shuttles.xlsx
layer: raw
preprocessed_companies:
type: pandas.CSVDataSet
filepath: data/02_intermediate/preprocessed_companies.csv
layer: intermediate
preprocessed_shuttles:
type: pandas.CSVDataSet
filepath: data/02_intermediate/preprocessed_shuttles.csv
layer: intermediate
model_input_table:
type: pandas.CSVDataSet
filepath: data/03_primary/model_input_table.csv
layer: primary
regressor:
type: pickle.PickleDataSet
filepath: data/06_models/regressor.pickle
versioned: true
layer: models
Run kedro-viz again with kedro viz
and observe how your visualisation has changed to indicate the layers:
Visualise charts in Kedro-Viz¶
Kedro-Viz supports Plotly and Matplotlib integrations so that you can share your data insights while exploring your pipeline.
Visualise Plotly charts in Kedro-Viz¶
Plotly is a free and open source Python library that allows you to make interactive, publication-quality graphs. With the Plotly integration on Kedro-Viz, you can output your interactive charts as part of your pipeline visualisation.
We have also used the Plotly integration to allow users to visualise metrics from experiments.
You must update the requirements.txt
file in your Kedro project and add the following datasets to enable Plotly for your project.
kedro[plotly.PlotlyDataSet, plotly.JSONDataSet]==0.18.3
You can view Plotly charts in Kedro-Viz when you use Kedro’s plotly datasets.
There are two types of Plotly datasets in Kedro: the plotly.PlotlyDataSet
and plotly.JSONDataSet
.
plotly.PlotlyDataSet
¶
To use this dataset, configure your plot in the catalog.yml
file. This dataset only supports Plotly Express.
Below is an example of how to visualise plots on Kedro-Viz using plotly.PlotlyDataSet
.
The below functions can be added to the nodes.py
and pipeline.py
files respectively.
import pandas as pd
def compare_passenger_capacity(preprocessed_shuttles: pd.DataFrame):
return preprocessed_shuttles.groupby(["shuttle_type"]).mean().reset_index()
def create_pipeline(**kwargs) -> Pipeline:
"""This is a simple pipeline which generates a plot"""
return pipeline(
[
node(
func=compare_passenger_capacity,
inputs="preprocessed_shuttles",
outputs="shuttle_passenger_capacity_plot",
),
]
)
Next, configure the plot in the catalog.yml
file:
shuttle_passenger_capacity_plot:
type: plotly.PlotlyDataSet
filepath: data/08_reporting/shuttle_passenger_capacity_plot.json
plotly_args:
type: bar
fig:
x: shuttle_type
y: passenger_capacity
orientation: h
layout:
xaxis_title: Shuttles
yaxis_title: Average passenger capacity
title: Shuttle Passenger capacity
plotly.JSONDataSet
¶
To use this dataset, configure your plot in your Kedro node. This dataset supports Plotly Express and Plotly Graph Objects.
Below is an example of how to visualise plots using Plotly Express and Plotly Graph Objects on Kedro-Viz using the plotly.JSONDataSet
.
The below functions can be added to the nodes.py
and pipeline.py
files respectively.
import plotly.express as px
import pandas as pd
# the below function uses plotly.express
def compare_passenger_capacity(preprocessed_shuttles: pd.DataFrame):
fig = px.bar(
data_frame=preprocessed_shuttles.groupby(["shuttle_type"]).mean().reset_index(),
x="shuttle_type",
y="passenger_capacity",
)
return fig
# the below function uses plotly.graph_objects
def compare_passenger_capacity(preprocessed_shuttles: pd.DataFrame):
data_frame = preprocessed_shuttles.groupby(["shuttle_type"]).mean().reset_index()
fig = go.Figure(
[
go.Bar(
x=data_frame["shuttle_type"],
y=data_frame["passenger_capacity"],
)
]
)
return fig
def create_pipeline(**kwargs) -> Pipeline:
"""This is a simple pipeline which generates a plot"""
return pipeline(
[
node(
func=compare_passenger_capacity,
inputs="preprocessed_shuttles",
outputs="shuttle_passenger_capacity_plot",
),
]
)
For plotly.JSONDataSet
, you must also specify the output type in the catalog.yml
file, like below.
shuttle_passenger_capacity_plot:
type: plotly.JSONDataSet
filepath: data/08_reporting/shuttle_passenger_capacity_plot.json
Once the above setup is completed, you can do a kedro run
followed by kedro viz
and your Kedro-Viz pipeline will show a new dataset type with icon . Click on the node to see a small preview of your Plotly chart in the metadata panel.
You can view the larger visualisation of the chart by clicking the ‘Expand Plotly Visualisation’ button on the bottom of the metadata panel.
Visualise Matplotlib charts in Kedro-Viz¶
Matplotlib is a Python library for creating static, animated, and interactive visualisations. Integrating Matplotlib into Kedro-Viz allows you to output your charts as part of your pipeline visualisation.
Note
The MatplotlibWriter dataset converts Matplotlib objects to image files. This means that Matplotlib charts within Kedro-Viz are static and not interactive, unlike the Plotly charts seen above.
You can view Matplotlib charts in Kedro-Viz when you use the Kedro MatplotLibWriter dataset. You must update the src/requirements.txt
file in your Kedro project by adding the following dataset to enable Matplotlib for your project:
kedro[matplotlib.MatplotlibWriter]==0.18.3
To use this dataset, configure your plot in your Kedro node. The below functions should be added to the nodes.py
and pipeline.py
files respectively.
# nodes.py
import matplotlib.pyplot as plt
def create_confusion_matrix(companies: pd.DataFrame):
actuals = [0, 1, 0, 0, 1, 1, 1, 0, 1, 0, 1]
predicted = [1, 1, 0, 1, 0, 1, 0, 0, 0, 1, 1]
data = {"y_Actual": actuals, "y_Predicted": predicted}
df = pd.DataFrame(data, columns=["y_Actual", "y_Predicted"])
confusion_matrix = pd.crosstab(
df["y_Actual"], df["y_Predicted"], rownames=["Actual"], colnames=["Predicted"]
)
sn.heatmap(confusion_matrix, annot=True)
return plt
# pipeline.py
def create_pipeline(**kwargs) -> Pipeline:
"""This is a simple pipeline which generates a plot"""
return pipeline(
[
node(
func=create_confusion_matrix,
inputs="companies",
outputs="dummy_confusion_matrix",
),
]
)
You must also specify the output type in the catalog.yml
file, like below. Remember to set the versioned flag to true
if you want to add the plots to experiment tracking as well.
reporting.dummy_confusion_matrix:
type: matplotlib.MatplotlibWriter
filepath: ${base_location}/08_reporting/dummy_confusion_matrix.png
versioned: true
Once this setup is completed, you can do a kedro run
followed by kedro viz
and your Kedro-Viz pipeline will show a new dataset node with this icon . Click on the node to see a small preview of your Matplotlib image in the metadata panel.
You can view the larger visualisation of the chart by clicking the ‘Expand Matplotlib Image’ button on the bottom of the metadata panel.