Create views
Creating, naming, and deleting views
To create a new view, use the New View button from the New View menu. Alternatively use shortcut ctrl-v (Mac)
or alt-v (Windows/Linux). To delete a view, use the bin icon in the Navigation Drawer.
To rename the title of the view or axes, simply click on the titles and type in your desired text.
Views from nodes
The Open View from Node function within the New View menu will open the best plot for your current selection. For example, if a numerical feature is selected, a histogram will open, whereas if a group of features are selected, a PCA plot of the dataset across these features will open. This is a fast way to quickly explore any feature. Moreover, if the nodes selected are themselves views then those views will be opened.
Creating Barplots
The distribution of any dataset across categorical features can be explored in Edge by first highlighting the
feature you wish to investigate, then navigating to Barplot from the New View section of the Menu Bar.
A new view will open containing the scenarios organised in a bar plot structure.
Creating Histograms
The distribution of a set of nodes according to their connections to a metric can be explored in Edge
by highlighting both the metric and set of nodes you wish to analyse, then navigating to Histogram
from the New View section of the Menu Bar.
Creating Scatter-plots
Select either 1 or 2 header nodes and then select Scatter plot from the New View menu. Read on to see how
you can also iteratively build a scatter plot by
assigning nodes to the x or y-axis directly.
Clustering Data using Embeddings
To cluster your dataset across multiple features, select the features you are interested in then select PCA embedding
or TSNE embedding from the New View menu.
What is an Embedding?
Both t-SNE (t-distributed Stochastic Neighbor Embedding) and PCA (Principal Component Analysis) are techniques used for dimensionality reduction in data visualization and analysis
PCA or t-SNE?
According to DataCamp, PCA preserves the variance in the data, whereas t-SNE preserves the relationships between data points in a lower-dimensional space.
- PCA is a linear dimensionality reduction technique. It is not ideal for capturing complex, non-linear relationships in the data. Use PCA when you have high-dimensional data and want to reduce the dimensionality while preserving as much variance as possible.
- t-SNE is a non-linear dimensionality reduction technique which is effective at revealing clusters and patterns in smaller datasets. It tends to preserve local structure well but may distort global structure, so it's important not to over-interpret global distances in the t-SNE space.
When generating either type of embedding, two views will appear. The first view presents all the data (row) nodes, organised based on their similarities across the input features used in the calculation. The second view, labeled "Feature Weights," displays the features themselves positioned to reflect how the data nodes were clustered. For instance, if Feature A is situated far to the left in the "Feature Weights" view, the data nodes associated with that feature are expected to be positioned farther to the left-hand side within the embedding view.
Creating a Label Embedding
To cluster nodes based on the semantic similarity of their labels, select the set of nodes whose labels
you want an embedding of and select Label Embedding from the New View menu.
Creating a Correlation View
To visualise the dependence between multiple variables in your taxonomy, utilize the correlation agent in the
New View menu. The view returned will contain a correlation matrix: the input variables will be plotted alongside new nodes
whose colour and label communicate the correlation coefficients between each pair of variables.
Important correlation values are highlighted by larger node sizes.
- Pearson Product-Moment is used to calculate the correlation between continuous (numerical) variables
- Theil's U is used to calculate categorical (nominal) correlations
- The Correlation ratio is used to generate the Numerical/ Categorical Correlation Ratios when a combination of variable types is provided as input to the agent
Duplicating Views
Views can be duplicated using the Duplicate View in the New View menu.
A new view will open automatically with the same nodes at the same positions as they were located in the original
view, but with new axis nodes. This means you can move nodes around in your new view without affecting the position
of nodes in the original view.
Organising Views
To re-order views, simply drag the view title towards the position you wish to assign to the view, then release.