Machine Learning
Clustering: COBWEB
COBWEB is an incremental provess for hierarchical conceptual clustering.
Every item has attributes (a1, a2, ..., an) in n categories. Each category can have several attributes.
First, define all categories. Then, select each category, and define all attributes of such category.
Once that is done, start creating items and defining their attributes in each of the categories.
Every item has attributes (a1, a2, ..., an) in n categories. Each category can have several attributes.
First, define all categories. Then, select each category, and define all attributes of such category.
Once that is done, start creating items and defining their attributes in each of the categories.
| Add Category | ||||
| Categories: | ||||
| Category Attributes: | ||||
| Add Attribute |
| Edit | Item Id |
| ＋ |
Solution
Calculate COBWEB hierarchical conceptual clustering from an input file and receive results in the file to be downloaded. Clustering hierarchy is a directed tree graph, where each node represents a cluster.
First, select the file format you wish to use:
CSV Input File Format:
CSV Output File Format:
JSON Input File Format:
JSON Output File Format:
Please select the file to upload:
First, select the file format you wish to use:
CSV Input File Format:
- File must be an ASCII file in a CSV format.
- First line must contain keyword [CATEGORIES] specifying that in the following lines comes the full list of the categories.
- Each category is specified on its own line, followed by the list of all of its attributes in the same line. Category and all attributes are separated by commas (',') and all whitespaces surrounding them are ignored, as are the empty lines.
- After all categories and their attributes are specified, next line must contain keyword [ITEMS] specifying that what follows in the consequent lines is the list of the items whose clustering we wish to find.
- Items are specified in tablular format where each row of a table is in a single line and specifies a single item. First column of the table must contain the name (id) of the item, and the rest of the columns contain the item's attributes corresponding to the list of the categories specified at the begining of the file.
- Column values (id and attributes) are separated by commas (',') and all the whitespaces surrounding them are ignored, as are the empty lines.
- Keywords are not case sensitive.
CSV Output File Format:
- Output file is an ASCII file in a CSV format named clusters.csv.
- Each cluster is specified by a keyword CLUSTER followed by its its name (id) in the next column. Top level cluster is named ROOT.
- Next row contains the keyword ITEMS, followed in the next column by the number of the items belonging to that cluster node. Items themselves are listed beneath, each in a sparate row.
- In order to prevent items named identically to create ambiguities, item names are appended by dot and the item's sequential number from the input, starting with 0.
- After all items are listed, next row contains the keyword CHILDREN, followed in the next column by the number of the children cluster nodes belonging to that cluster node. Each child cluster is then listed beneath, but by one coulumn to the right.
JSON Input File Format:
- File must be an ASCII file containing a single JSON object.
- That object must conatin two members: "categories" and "items" (both lower case).
- Value of the categories member is an array whose elements are objects describing individual categories:
- Each category object must contain two members: "category" and "attributes" (both lower case).
- category memeber contains the name of the category (string).
- attributes memeber contains the the array containig all possible attributes that category can take.
- Value of the items member is an array whose elements are arrays describing individual items:
- Each item is representedas an array.
- First element or the aray is the name (id) of the item.
- All other elements of the array are item's attributes corresponding categories from the categoies array.
JSON Output File Format:
- Output file is an ASCII file named clusters.json, containing a single JSON object.
- That object represents the top level (root) node of the clustering hierarchy tree.
- Each node in the tree represents the cluster. It is represented as a JSON object containing the following 3 members:
- name - containg the name/id of the cluster. Name of the root node is "ROOT".
- items - containg the JSON array whose elements are the items belonging to that cluster. ROOT node will not have eny items belonging to it, so its items array will be an empty array.
- children - containg the JSON array whose elements are the children nodes (clusters) of that cluster. Leaf nodes (clustes) of the clustering tree will have null for the value of the children.
Result file has been successfully created.