Visualize and classify complex multidimensional datasets with only two clicks. The app can be used in combination with my django_machine_learning_supervised_classification repository. In contrast to the latter, the project here is build with Django's Class Based Views.
Load unclassified datasets from your local machine or remote resources via URL.
The dataset is then projected in two dimensions via Principal Component Analysis (PCA). A plotly express scatter plot will be displayed as in the following snapshot. Underneath you find the head of the data table you just loaded. Type the number of clusters (or increment by arrows) that you expect, meaning how many clusters do you see in the plot. Submit again.
In the following, I will roughly explain how the views.py works, for a detailed explanation please read the comments in the file itself. The views.py consists of two parts, one that processes your data set and one for the actual views. Throughout both parts, classes inherit from top to bottom.
The first part has three classes. The first step is to read in a dataset which is achieved by the class ClusterBaseData. The dataset is then saved as a pickle object in the database for future analysis. The DimReduction class projects the multidimensional data into two dimensions via PCA and saves both, a data table and a plot of the new 2D dataset. This class is applied twice, for the raw (yet unclassified) data and the classified data. The latter will be delivered by the class Cluster, which applies the k-Means_method to classify the DimReduction objects.
The second part of views.py contains two views. THe IndexView inherits from FormView and TemplateView to process the input form and the results at the same time. It also inherits all methods of the data processing part by the class Clusters. Initially, only the form is displayed. On submitting the URL, the 2D plot and the head of the data table of the original nonclassified data are shown. If then the number of expected clusters is submitted the class ResultsView is called which has basically the same structure but processes the clustered data instead of the nonclassified.
The following graph illustrates the data flow within the views.py. circles, squares, and triangular represent variables, classes and templates respectively. Red, violet and green distinguish between classes from models.py, data processing classes and views.
If a dataset has already beíng processed by the app, the same calculation doesn't have to be done over and over again. This is why all the results are saved as instances of two classes in the database. The attributes of the corresponding classes also serve as identifiers. Data frames and plots are stored as a PickledObjectField(). The class BaseData contains the original nonclassified dataset. The class ProjectionIn2D contains the PCA projection of both, the original (nonclassified) and the clustered (by k-Means classified) dataset.
