Systems and methods for reduction and interpretability of continuous action probability trees

Abstract

A system performs a method for improving the efficiency of operating a device. A processor obtains a first decision tree available for device operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, where the first child node and the second child node are accessible from the parent node of the first layer. The processor selects the parent node, identifies and clusters the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature common between the first child node and the second child node, and determines a semantic meaning of the clustered child nodes that is not present in the first child node or the second child node.

Description

Technical Field

[0001] The subject disclosure relates to operation of autonomous vehicles using decision trees, and more particularly, to systems and methods for reducing and adding partial semantic meaning to a decision tree prior to use at an autonomous vehicle. Background Art

[0002] Autonomous or semi-autonomous vehicles can execute and use decision trees, which provide a logical framework for selecting the next action for the vehicle. The decision tree relates the vehicle's current state to multiple possible future states. An algorithm is used to traverse the decision tree. As the algorithm traverses the decision tree, the computational effort increases with the tree's size. Therefore, the use of algorithms for large decision trees is reduced in real-time scenarios. Furthermore, the actions may lack semantic meaning, hindering the ability of the algorithm's selected actions to be understood. Therefore, it is desirable to provide a method for reducing decision trees and increasing semantic meaning while preserving information, allowing for more computationally cost-effective implementation at the vehicle. Summary of the Invention

[0003] In one exemplary embodiment, a method for improving the efficiency of operating a device is disclosed. The method includes: obtaining a first decision tree usable in device operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, wherein the first child node and the second child node are accessible from the parent node of the first layer; selecting the parent node and identifying at least the first child node and the second child node; clustering the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature common between the first child node and the second child node; and determining semantic meanings of the clustered child nodes, wherein the semantic meanings of the clustered child nodes are not present in the first child node or the second child node.

[0004] In addition to one or more features described herein, determining semantic meaning further comprises ranking at least one feature based on an amount of contribution of the at least one feature to the clustering.

[0005] In addition to one or more features described herein, the method further includes selecting a distinguishing feature from the at least one feature that contributes most to the clustering, and assigning semantic meaning to the cluster subnodes based on the distinguishing feature.

[0006] In addition to one or more features described herein, wherein the first decision tree includes a third layer having nodes, each node of the third layer being accessible from a node of the second layer, the method further includes designating a cluster child node as a cluster parent node, identifying nodes of the third layer that are related to the cluster parent node, and clustering the identified nodes of the third layer.

[0007] In addition to one or more features described herein, the method also includes operating a device using the first decision tree to take an action based on one of the first child node and the second child node, and presenting a reason for the action based on semantic meaning of clustered child nodes from the second decision tree.

[0008] In addition to one or more features described herein, the at least one feature includes at least one of a device state, an action value, and a spatial parameter of the action.

[0009] In addition to one or more features described herein, the parent node is a top node of the first decision tree.

[0010] In another exemplary embodiment, a system for improving the efficiency of device operation is disclosed. The system includes a processor. The processor is configured to: obtain a first decision tree usable in device operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, wherein the first child node and the second child node are accessible from the parent node of the first layer; select the parent node and identify at least the first child node and the second child node; cluster the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature common between the first child node and the second child node; and determine a semantic meaning of the clustered child nodes, wherein the semantic meaning of the clustered child nodes is not present in the first child node or the second child node.

[0011] In addition to one or more features described herein, the processor is further configured to determine semantic meaning by ranking at least one feature based on an amount that the at least one feature contributes to the cluster.

[0012] In addition to one or more features described herein, the processor is further configured to select a distinguishing feature that contributes most to the clustering from the at least one feature, and assign semantic meaning to the cluster sub-nodes based on the distinguishing feature.

[0013] In addition to one or more features described herein, the first decision tree includes a third layer having nodes, each node of the third layer being accessible from a node of the second layer, and wherein the processor is further configured to designate a cluster child node as a cluster parent node, identify nodes of the third layer that are related to the cluster parent node, and cluster the identified nodes of the third layer.

[0014] In addition to one or more features described herein, the processor is further configured to operate the device using the first decision tree to take an action based on one of the first child node and the second child node, and to present a reason for the action based on a semantic meaning of the clustered child nodes from the second decision tree.

[0015] In addition to one or more features described herein, the at least one feature includes at least one of a device state, an action value, and a spatial parameter of the action.

[0016] In addition to one or more features described herein, the parent node is a top node of the first decision tree.

[0017] In another exemplary embodiment, a vehicle is disclosed. The vehicle includes a processor. The processor is configured to: obtain a first decision tree usable in vehicle operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, wherein the first child node and the second child node are accessible from the parent node in the first layer; select the parent node and identify at least the first child node and the second child node; cluster the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature shared between the first child node and the second child node; and determine a semantic meaning of the clustered child nodes, wherein the semantic meaning of the clustered child nodes is not present in either the first child node or the second child node.

[0018] In addition to one or more features described herein, the processor is further configured to determine semantic meaning by ranking at least one feature based on an amount by which the at least one feature contributes to the cluster.

[0019] In addition to one or more features described herein, the processor is further configured to select a distinguishing feature that contributes most to the clustering from the at least one feature, and assign semantic meaning to the cluster sub-nodes based on the distinguishing feature.

[0020] In addition to one or more features described herein, wherein the first decision tree includes a third layer having nodes, each node of the third layer being accessible from a node of the second layer, the processor is further configured to designate a cluster child node as a cluster parent node, identify nodes of the third layer that are related to the cluster parent node, and cluster the identified nodes of the third layer.

[0021] In addition to one or more features described herein, the processor is further configured to operate the device using the first decision tree to take an action based on one of the first child node and the second child node, and to present a reason for the action based on a semantic meaning of the clustered child nodes from the second decision tree.

[0022] In addition to one or more features described herein, the at least one feature includes at least one of a device state, an action value, and a spatial parameter of the action.

[0023] The above features and advantages and other features and advantages of the present disclosure will become apparent when the following detailed description is read in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Additional features, advantages, and details appear by way of example only in the following detailed description, which refers to the accompanying drawings, in which:

[0025] Figure 1 A vehicle capable of operating in an autonomous or automatic mode is shown according to an exemplary embodiment;

[0026] Figure 2 shows a portion of a decision tree that may be used to operate a vehicle in an illustrative embodiment;

[0027] Figure 3 A flow chart showing a method of operating a vehicle using a decision tree in an illustrative embodiment;

[0028] Figure 4 A data set showing illustrative nodes of an original decision tree in an illustrative embodiment;

[0029] Figure 5 In the illustrative embodiment shown Figure 2 A diagram of a decision tree of FIG, wherein the labeling shows the first iteration of the tree reduction process;

[0030] Figure 6 is a diagram showing a second iteration of the tree reduction process in an illustrative embodiment;

[0031] Figure 7 Shown by Figure 5 and Figure 6 The intermediate decision tree formed by the tree reduction process shown;

[0032] Figure 8 shows a flow chart of a tree compression method disclosed herein in an illustrative embodiment;

[0033] Figure 9 A data set showing cluster nodes of a reduced decision tree in an illustrative embodiment; and

[0034] Figure 10 A view of a vehicle instrument panel is shown in an illustrative embodiment. DETAILED DESCRIPTION

[0035] The following description is merely exemplary in nature and is not intended to limit the present disclosure, its application, or uses. It should be understood that throughout the drawings, corresponding reference numerals indicate the same or corresponding parts and features.

[0036] According to an exemplary embodiment, Figure 1A vehicle 100 is shown that can operate in an autonomous or automatic mode. Vehicle 100 can be a fully autonomous vehicle or a semi-autonomous vehicle. Vehicle 100 includes a drive system 102 that controls the vehicle's autonomous operation. Drive system 102 includes a sensor system 104 for obtaining information about the vehicle's surroundings or environment, a controller 106 for calculating possible actions for the autonomous vehicle based on the obtained information and for implementing one or more possible actions, and a human-machine interface 108 for explaining the actions taken by the vehicle to a vehicle occupant (e.g., a driver or passenger). Human-machine interface 108 need not be dedicated solely to explaining actions and can be an interface within the vehicle for purposes other than explaining actions. Sensor system 104 can include devices such as cameras, lidar, radar, GPS, etc. Controller 106 can include processing circuitry that may include an application-specific integrated circuit (ASIC), electronic circuitry, a processor (shared, dedicated, or grouped) executing one or more software or firmware programs and memory, combinational logic circuitry, and / or other suitable components that provide the described functionality. According to one or more embodiments detailed herein, controller 106 may include a non-transitory computer-readable medium storing instructions that, when processed by one or more processors of controller 106, implement a method for reducing a decision tree, adding semantic meaning to the reduced decision tree, and using the reduced decision tree at vehicle 100 to provide a message explaining an action to a vehicle occupant based on the added semantic meaning. Controller 106 may operate a program that executes artificial intelligence. Human-machine interface 108 may include one or more interfaces, such as a speaker, a display, a tactile device, and the like. Each interface is adapted for a given message format. For example, a speaker may be used to present auditory information. A display may be used to present written information or a chart. A tactile device may be used to present tactile information, such as a vibration felt by the occupant. In another embodiment, human-machine interface 108 may be a mobile device that the occupant can use while inside or outside the vehicle.

[0037] The vehicle 100 also includes a communication unit 110 capable of communicating with a remote processor 120. In various embodiments, the processor of the controller 106 and / or the remote processor 120 may execute the methods disclosed herein for tree reduction and adding semantic meaning to decision trees, as disclosed herein.

[0038] Figure 2A portion of a decision tree 200 that can be used by controller 106 to operate vehicle 100 in an illustrative embodiment is shown. Decision tree 200 is a data structure comprising multiple nodes organized along branches. Edges connect any two nodes of the decision tree and can be parameterized by the probability of moving from one node to the connected node. Each node represents a state of vehicle 100, and each edge represents an action that takes the vehicle from one state to another (a subsequent state or a future state). Decision tree 200 includes multiple layers. Each layer includes one or more nodes of decision tree 200. While decision tree 200 can have any number of layers, for illustrative purposes, a first layer 202, a second layer 204, a third layer 206, and a fourth layer 208 are shown. First layer 202 includes a top node 210 representing the current state of the vehicle. Second layer 204 includes nodes 212a-212i accessible to top node 210. Third layer 206 includes nodes 220a-220h accessible to nodes 212a-212i of second layer 204. This continues through the remaining layers of the decision tree (not shown). Nodes in successive layers can form a parent-child relationship, where nodes in higher layers (i.e., layers with lower layer numbers) are referred to as parent nodes, and nodes in lower layers (i.e., layers with higher layer numbers) are referred to as child nodes. Thus, for example, top node 210 is the parent node of nodes 210a-210i in the second layer, and nodes 210a-210i are the child nodes of the top node. In a parent-child relationship, a parent can have many children, but a child can have only one parent.

[0039] The controller 106 selects an action to be performed from the root node of the tree as the best action (where "best" can be based on the number of visits or some other criteria). Once the vehicle performs the action, a new tree can be constructed based on the new current state of the vehicle. The construction of these trees can be implemented as in a Monte Carlo tree search algorithm.

[0040] Figure 3 A flow chart 300 is shown of a method for operating a vehicle using an original or first decision tree in an illustrative embodiment. The method includes a reduction operation 304 for generating a reduced or second decision tree from the first decision tree. The second decision tree can be used to perform additional calculations on the first decision tree or for comparisons between successive trees. Using the second decision tree (instead of the first decision tree) reduces computation time and computational cost. The reduction operation 304 can be performed offline or online. The reduction is performed after the first decision tree has been constructed. A first decision tree 302 (e.g., Figure 2Decision tree 200). The first decision tree 302 is sent through a reduction operation 304, which includes two processes performed in parallel. The first process 306 is a clustering and unification process, and the second process 308 is an interpretability process. The clustering and unification process reduces the first decision tree to a second decision tree (reduced decision tree) by forming cluster nodes within one or more layers of the tree. Based on the commonality of node features, the nodes within a layer are clustered and unified to form cluster nodes within the layer. The child nodes of the cluster nodes are then identified (in subsequent layers) and clustering is performed between the child nodes. The interpretability process determines the semantic meaning of the cluster nodes. The reduction operation 304 produces a second decision tree 310. When an action is taken to select a cluster node of the second decision tree 310, the semantic meaning from the selected cluster node is displayed at the human-computer interface 312.

[0041] In more detail, a tree reduction process receives a first decision tree. The first decision tree includes at least a first layer having at least one parent node and a second layer having at least first and second child nodes. The first and second child nodes can be accessed from the parent node of the first layer by taking an action. A parent node is selected and child nodes (e.g., at least the first and second child nodes) are identified. At least the first and second child nodes are clustered together to form cluster child nodes of a second decision tree, wherein the cluster child nodes are based on at least one common feature between the first and second child nodes, thereby generating cluster child nodes that include at least the first and second child nodes. Semantic meaning is determined for the cluster child nodes and assigned to the child nodes. The semantic meaning of the cluster child nodes can be a meaning that is not present in any individual node that is clustered to form the cluster node. The result of these processes is a reduced decision tree. Using the cluster child nodes of the reduced decision tree, additional computations can be performed on the reduced (second) decision tree. An action for the vehicle is selected by implementing the first decision tree to operate the vehicle. Reasons for the actions are presented based on the corresponding cluster groups in the second decision tree to achieve more efficient computation and accessible semantic meaning.

[0042] Figure 4 A data set 400 of illustrative nodes of a first decision tree 302 in an illustrative embodiment is shown. The data set includes various features of actions, including the action taken 402 and the number of visits 404 at the node. These features can be used in the clustering process. The features can also include state information, a value component of the action, and spatial parameters of the action. The state information includes information about the resulting state of the vehicle or other entities around the vehicle as a result of the action. The state information can include, but is not limited to, the distance of the other entities from the vehicle, their relative speed, and position. The value component represents the value of the action as it relates to the overall driving experience. The value component relates not only to the current action, but also to possible future actions. Exemplary value components are shown in FIG. Figure 4The spatial parameters describe the parameters of the action taken at the node and may include features such as the latitude 408 of the vehicle, the longitude of the vehicle, the acceleration 410 of the vehicle along the trajectory, etc.

[0043] Node clustering is performed using features found in the dataset of nodes. For example, clustering can be based on vehicle speed, vehicle acceleration, trajectory curvature, etc. Methods for clustering nodes include applying a clustering algorithm to the node features. Exemplary clustering algorithms include, but are not limited to, k-means clustering, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise), etc.

[0044] Figure 5 is a diagram showing an illustrative embodiment. Figure 2 500 of a decision tree 200, wherein the first iteration of the tree reduction process is shown. A parent node is selected within the decision tree, and the child nodes of the parent node are clustered according to their characteristics. Although the node selected as the parent node can be any node of the decision tree, the top node 210 is usually selected to start the reduction process. Starting from the top node 210, child nodes (i.e., nodes 212a-212i) are identified within the second layer 204. The clustering process is performed on the child nodes (e.g., by nodes 212a-212i). As a result of clustering, nodes 212a, 212b, 212d, and 212e are grouped together to form a first cluster node 502 based on their feature values. In addition, nodes 212f and 212g are grouped together to form a second cluster node 504 based on their feature values. In addition, nodes 212c, 212h, and 212i are grouped together to form a third cluster node 506 based on their feature values.

[0045] Figure 6600 is a diagram illustrating the second iteration of the tree reduction process in an illustrative embodiment. The second iteration comprises performing a clustering process on the next layer of decision tree (e.g., at the third layer 206). The clustering nodes of the second layer 204 can now be labeled as clustering parent nodes. The child nodes of each clustering node are identified (e.g., the first clustering node 502, the second clustering node 504, and the third clustering node 506). For the first clustering node 502, the child nodes are identified as node 220a (child node 212a of node), node 220b (child node of node 212b), node 220c (child node of node 212d), and node 220d (child node of node 212e). These nodes are unified into the first brother group 602. For the second clustering node 504, the child nodes are identified as node 220e (child node of node 212f) and node 220f (child node of node 212g). These nodes are unified into the second brother group 604. For the third cluster node 506, the child nodes are identified as node 220g (child of node 212h) and node 220h (child of node 212i). These nodes are united into a third sibling group 606. Clustering at this level is performed separately for each sibling group.

[0046] Figure 7 Shown from Figure 5 and 6 The tree reduction process shown forms an intermediate decision tree 700. The intermediate decision tree includes the first cluster node 502, the second cluster node 504, the third cluster node 506, and any other cluster nodes of the second layer 204. The edges from the decision tree 200 are replaced by the appropriate reduction edges of the reduced decision tree. For example, the parent node 201 is replaced with nodes 212, 212b, 212d, and 212e ( Figure 5 ) are replaced by reduced edges 702. Nodes 220a, 220b, 220c, and 220d, which are children of nodes 212a, 212b, 212d, and 212e, respectively, are now shown under the first cluster node 502 in the first sibling group 602. Nodes 220e and 220f, which are children of nodes 212f and 212g, respectively, are shown under the second cluster node 504 in the second sibling group 604. Nodes 220g and 220h, which are children of nodes 212h and 212i, respectively, are shown under the third cluster node 506 in the third sibling group 606.

[0047] Using the intermediate decision tree 700, a clustering method can be used to cluster the nodes of the second layer 204 within the same sibling group. For example, a first clustering is performed on the nodes of the first sibling group 602 (e.g., nodes 220a-220d), a second clustering is performed on the nodes of the second sibling group 604 (e.g., nodes 220e and 220f), a third clustering is performed on the nodes of the third sibling group 606 (e.g., nodes 220g and 220g), and so on.

[0048] Figure 8 A flowchart 800 of the tree reduction method disclosed herein in an illustrative embodiment is shown. In block 802, using the original decision tree, a parent cluster node is obtained and its children are identified. For the first iteration, the parent cluster node is the top node 210 of the decision tree. In block 804, the children of the parent cluster node are clustered using a suitable clustering method. In block 806, an interpretability process is performed on the clustered children. In block 808, the clustered children are unified into cluster nodes. In block 810, a check is performed to determine whether the last level of the decision tree (or a particular branch of the decision tree) has been reached. If the last level has been reached, the method proceeds to block 812, where the method ends. The result is a reduced decision tree. Returning to block 810, if the last level has not been reached, the method proceeds to block 814. In block 814, the cluster child node is reassigned as a new parent cluster node. The children of the new parent cluster node are identified and collected, and the method returns to block 802.

[0049] Figure 9 A data set 900 of cluster nodes of a reduced decision tree in an illustrative embodiment is shown. The data set 900 includes an action 902 taken at the cluster node, a number of visits 904, and a value 906 of the action. The data set 900 of the cluster nodes also shows various features of the action, such as spatial parameters, value components, and state information. For illustrative purposes, trajectory acceleration 908 and latitude 910 are shown. The data set 900 also includes a record of the main features 912 used in the clustering process of creating cluster nodes. For illustrative purposes, the main features include velocity (VEL), the curvature value of the trajectory (CURVATURE), and the acceleration at a selected point of the trajectory (Traj_accel). Traj_accel refers to the acceleration of the 7th point of the trajectory (a point out of the 10 points that define the trajectory). The interpretability process organizes the features from most important to least important (relative to the clustering process). The most important features can be considered as the distinguishing features of the cluster nodes.

[0050] Various interpretability methods can be used. As an example, the interpretability problem can be transformed into a binary supervised clustering problem, for which decision trees (DT) or random forests (RF) provide one or more preferential discriminative features.

[0051] Figure 10A view 1000 of the instrument panel of vehicle 100 in an illustrative embodiment is shown. When an action is taken on vehicle 100, a message may be displayed on a display 1002 on the instrument panel. The message explains to the occupants of vehicle 100 the reason for taking the action. The reason may be based on the action taken by the vehicle, the characteristics of the action taken by the vehicle, the value of the action taken by the vehicle, the state of the vehicle, or a combination thereof. Alternatively, the reason may be a comparative explanation, explaining why an action was taken that was opposite to an action not taken, or as a benefit relative to an action not taken. The reason may be based on distinguishing features from cluster nodes. For example, the message may state, "We chose to perform this action because the speed and longitudinal jerk of this action performed better than the other action" or "We chose a stronger deceleration because we achieved a greater headway," etc.

[0052] The terms "a" and "an" do not indicate a limitation of quantity, but rather indicate the presence of at least one of the referenced item. The term "or" means "and / or" unless the context clearly indicates otherwise. References to "an aspect" throughout this specification mean that a particular element (e.g., feature, structure, step, or characteristic) described in conjunction with that aspect is included in at least one aspect described herein and may or may not be present in other aspects. Furthermore, it should be understood that the described elements may be combined in any suitable manner in various aspects.

[0053] When an element such as a layer, film, region, or substrate is referred to as being "on" another element, it can be directly on the other element or intervening elements may also be present. In contrast, when an element is referred to as being "directly on" another element, there are no intervening elements present.

[0054] Unless otherwise indicated herein, all test standards are the most recent standards in effect as of the filing date of this application or, if priority is claimed, the filing date of the earliest priority application in which the test standards appear.

[0055] Unless defined otherwise, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0056] Although the above disclosure has been described with reference to exemplary embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the scope thereof. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the present disclosure without departing from the essential scope of the present disclosure. Therefore, it is intended that the present disclosure is not limited to the particular embodiments disclosed, but is intended to include all embodiments falling within its scope.

Claims

1. A method for improving the efficiency of operating equipment, comprising: obtaining a first decision tree usable in device operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, wherein the first child node and the second child node are accessible from the parent node of the first layer; selecting a parent node and identifying at least a first child node and a second child node; clustering the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature common between the first child node and the second child node; as well as A semantic meaning of the cluster child node is determined, wherein the semantic meaning of the cluster child node is not present in the first child node or the second child node.

2. The method according to claim 1, wherein Determining the semantic meaning further includes ranking at least one feature based on an amount by which the at least one feature contributes to the clustering. 3 . The method according to claim 2 , further comprising selecting a distinguishing feature that contributes most to clustering from the at least one feature, and assigning the semantic meaning to the cluster sub-node based on the distinguishing feature.

4. The method according to claim 1, wherein The first decision tree includes a third layer having nodes, each node of the third layer is accessible from a node of the second layer, and the method further includes designating the cluster child node as a cluster parent node, identifying nodes of the third layer that are related to the cluster parent node, and clustering the identified nodes of the third layer.

5. The method of claim 1 further comprising operating the device using the first decision tree to take an action based on one of the first and second child nodes, and presenting a reason for the action based on semantic meaning of clustered child nodes from the second decision tree.

6. A system for improving the efficiency of equipment operation, comprising: Processor, configured as: obtaining a first decision tree usable in device operation, the first decision tree having a first layer including a parent node and a second layer including at least a first child node and a second child node, wherein the first child node and the second child node are accessible from the parent node of the first layer; selecting a parent node and identifying at least a first child node and a second child node; clustering the first child node and the second child node to form a second decision tree having clustered child nodes, the clustered child nodes being based on at least one feature common between the first child node and the second child node; as well as A semantic meaning of the cluster child node is determined, wherein the semantic meaning of the cluster child node is not present in the first child node or the second child node.

7. The system according to claim 6, wherein: The processor is further configured to determine the semantic meaning by ranking at least one feature based on an amount by which the at least one feature contributes to a cluster.

8. The system according to claim 7, wherein: The processor is further configured to select a distinguishing feature that contributes most to clustering from the at least one feature, and assign the semantic meaning to the cluster sub-node based on the distinguishing feature.

9. The system according to claim 6, wherein: The first decision tree includes a third layer having nodes, each node of the third layer is accessible from a node of the second layer, and wherein the processor is further configured to designate the cluster child nodes as cluster parent nodes, identify nodes of the third layer that are related to the cluster parent nodes, and cluster the identified nodes of the third layer.

10. The system according to claim 6, wherein: The processor is further configured to operate the device using the first decision tree to take an action based on one of the first and second child nodes, and to present a reason for the action based on semantic meaning of clustered child nodes from the second decision tree.

Images