A cascading defect cluster classification method based on material numerical calculation big data

Abstract

The application discloses a cascaded defect cluster classification method based on material numerical calculation big data, and comprises the following specific steps: S1: obtaining three-dimensional coordinates: obtaining three-dimensional coordinates of each point defect in a cluster; S2: calculating distance and angle: calculating the distance and angle between adjacent point defects; S3: counting number: mapping the calculated distance and angle results to different intervals, and counting the number of each interval; S4: calculating a defect cluster feature vector: describing the cluster geometry based on the distance and angle of adjacent point defects in the cluster, and calculating the feature vector of each cluster by using the three-dimensional space coordinates of each point defect. The cascaded defect cluster classification method based on material numerical calculation big data has the effect of solving the problem that the cluster information characteristic description with large noise and low flexibility influences the classification effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of material irradiation effect and the field of data mining, and in particular to a cascade defect cluster classification method based on material numerical calculation big data. BACKGROUND

[0002] Nuclear reactor materials are bombarded by high-energy particles, which will cause cascade collisions and produce irradiation damage. Material numerical calculation big data generated by numerical nuclear reactor simulation contains a large amount of irradiation damage information under different calculation scales and service environments. They all belong to calculation data related to material irradiation damage, and are closely related to each other. Through the study of the correlation between these data, further mining and analysis of important physical information in the data can obtain new understanding of nuclear reactor material irradiation effect and material mechanism, so as to present the value contained in material numerical calculation big data in a more clear and intuitive form.

[0003] Defect cluster structures formed by cascade collisions play an important role in the microstructure evolution during irradiation. In order to promote the research of nuclear material irradiation damage, it is necessary to classify the defect clusters in terms of geometric shape to obtain different cluster shapes and explore the relationship between cluster type and number and experimental environment. In the related research of describing cluster structure in three-dimensional space coordinates, the geometric shape of cluster space structure is often affected by space translation, rotation and scaling, so that clusters with similar geometric shapes are divided into different cluster categories in space. Therefore, a method is needed to extract features from coordinate data to maximize the translational and rotational invariance of clusters. At the same time, material numerical calculation big data obtained by numerical simulation has the characteristics of large data volume, and the feature vectors extracted from the data will also have the problem of high dimension, which is easy to cause the problem of curse of dimensionality. Since the result data after cascade collision is usually the coordinate data of related atoms, it is difficult to intuitively find the information value it contains, such as cluster type and cluster classification result, so it is necessary to use clustering algorithm to classify them and visualize them.

[0004] The present application provides a defect cluster clustering method based on material numerical calculation big data to solve the problem of low flexibility of cluster information feature description affecting classification effect. SUMMARY

[0005] The present application discloses a cascade defect cluster classification method based on material numerical calculation big data, which aims to solve the technical problem of low flexibility of cluster information feature description affecting classification effect.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme:

[0007] A cascade defect cluster classification method based on material numerical calculation big data, comprising the following specific steps:

[0008] S1: Obtain three-dimensional coordinates: obtain the three-dimensional coordinates of each point defect in the cluster;

[0009] S2: Calculate the distance and angle: calculate the distance and angle between adjacent point defects;

[0010] S3: Statistics: map the calculated distance and angle results to different intervals and count the number of each interval;

[0011] S4: Calculate the defect cluster feature vector: describe the cluster geometry based on the distance and angle between adjacent point defects in the cluster, and calculate the feature vector of each cluster using the three-dimensional spatial coordinates of each point defect;

[0012] S5: Dimension reduction: use the UMAP algorithm to reduce the high-dimensional feature vector obtained to obtain the low-dimensional structure between data;

[0013] S6: Clustering: use the density-based clustering algorithm HDBSCAN to visualize the cluster classification results, and use the silhouette coefficient as an evaluation index of clustering effect;

[0014] S7: Visualize the cluster classification results: process the cluster classification results to make them visual.

[0015] By describing the cluster geometry based on the distance and angle between adjacent point defects in the cluster, calculating the feature vector of each cluster using the three-dimensional spatial coordinates of each point defect, and using the UMAP algorithm to reduce the high-dimensional feature vector obtained to obtain the low-dimensional structure between data, the cascade collision defect cluster of the material irradiation is identified and classified, and the evolution mechanism of the material defect cluster is mined. The relationship between the experimental environment and the characteristics and properties of the cluster is established, so that the cluster classification method can simply and flexibly represent the cluster geometry and has high calculation efficiency, eliminates the noise caused by the same geometric shape cluster under the influence of translation, scaling and rotation, qualitatively determines the global and local relationship between clusters, analyzes the relationship between the experimental environment and the characteristics and properties of the cluster, lays a foundation for mining and discovering the properties of the cluster and its influence, and solves the problem of cluster information feature description affecting classification effect due to large noise and low flexibility.

[0016] In a preferred scheme, in S4, calculating the defect cluster feature vector comprises the following specific steps:

[0017] S41: Store the feature vector: create distance_array and angle_array to store the distance feature vector and angle feature vector, respectively;

[0018] S42: Calculate the distance between adjacent defect atoms: traverse the point defects in the cluster, calculate the distance between two adjacent defect atoms, and the angle of the adjacent defect atoms relative to the third defect atom;

[0019] S43: Divide by the maximum distance: divide each distance by the maximum distance, and the distance range is 0≤d≤1 at this time;

[0020] In the S4, the calculation of the defect cluster feature vector includes the following specific steps:

[0021] S44: Process the number of interval distances: divide m intervals with the maximum distance, map each distance to the corresponding interval, count the number of distances falling into each interval, and perform normalization processing to obtain an m-dimensional distance feature vector;

[0022] S45: Divide by the interval angle: divide each angle by the interval angle α, where the interval angle is set according to the number of intervals n to be divided, count the number of angles falling into each interval, and perform normalization processing to obtain an n-dimensional angle feature vector;

[0023] S46: Construct the defect cluster feature vector: the m-dimensional distance feature vector and the n-dimensional angle feature vector calculated by the above steps constitute the m+n-dimensional feature vector of the defect cluster, and jump to dimensionality reduction;

[0024] In the S42, the angle formed by the two defect atoms relative to the third defect atom is in the range of 0≤θ≤180°.

[0025] By dividing each distance by the maximum distance, the influence of cluster shape rotation and scaling in space on cluster geometric features can be eliminated, ensuring the stability of cluster geometric features. At the same time, by taking the three-dimensional space coordinates of each point defect in the defect cluster as input, the geometric shape is characterized, and the high-dimensional feature space is processed with dimensionality reduction under the premise of preserving the original data information to the greatest extent.

[0026] In a preferred scheme, the S5 includes the following specific steps:

[0027] S51: Set parameters: set the parameters n_components, n_neighbors, and min_dist in the UMAP algorithm to achieve different dimensionality reduction effects;

[0028] S52: Construct a neighbor graph: assuming that the m+n high-dimensional feature vector data samples of the cluster are uniformly distributed in the topological space, use the KNN algorithm to find the nearest neighbor, and construct a weighted K-neighbor graph of the feature vector distribution of each point defect in the cluster;

[0029] S53: execute the algorithm: map the obtained high-dimensional cluster feature vector to a low-dimensional space, execute the UMAP algorithm, and realize dimension reduction;

[0030] In the parameter setting S51, the parameter n_components is the dimension of the cluster feature space after dimension reduction, and n_components is set to 2 to reduce the cluster feature space to two dimensions for direct visualization and distinguish cluster types. The parameter n_neighbors is the number of nearest neighbors, and a smaller value is mainly based on the local structure of the cluster for division, and a larger value tends to the overall structure of the cluster. The min_dist is the minimum distance between the high-dimensional feature vector space of each cluster and the low-dimensional space point after dimension reduction;

[0031] The S6 cluster includes the following specific steps:

[0032] S61: Set the parameter clustering effect: set the parameters min_cluster_size and min_samples in the HDBSCA algorithm;

[0033] S62: Execute the HDBSCAN algorithm: execute the HDBSCAN algorithm according to the low-dimensional space feature vector obtained after dimension reduction, perform clustering, and obtain the cluster class label after classification;

[0034] S63: Calculate the clustering result: input the low-dimensional space feature vector and label to calculate the clustering result, and take the silhouette coefficient as the evaluation index of cluster clustering. The value range of the silhouette coefficient is [-1, 1]. The closer the distance between the cluster samples with similar geometric shapes, the farther the distance between the cluster samples with different geometric shapes, and the higher the value.

[0035] S64: Determine the silhouette coefficient value: if the silhouette coefficient value is not ideal, jump to S5, the dimension reduction step, and re-set the parameters until the clustering effect reaches the ideal effect;

[0036] In the parameter setting S61, the parameter min_cluster_size is the number of clusters contained in each cluster class, and the larger the parameter value, the fewer the cluster classes. The parameter min_samples is the number of clusters contained in the neighborhood of a point, and the larger the parameter value, the more conservative the clustering, and more points are considered as noise.

[0037] The local structure of the cluster is mainly divided by setting a smaller value in the parameter n_neighbors, and a larger value tends to the overall structure of the cluster, thereby losing the accuracy of the local structure details. By appropriately setting the size of the parameter, the local structure and the global structure of the cluster are balanced, so that the subsequent clustering effect can be improved. The min_dist set by the parameter is the minimum distance between the points in the low-dimensional space obtained by reducing the high-dimensional feature vector space of each cluster. The lower the value, the closer the points embedded in the low-dimensional space, and the more attention is paid to the local structure of the cluster. A larger value will pay attention to the global structure of the cluster, and more noise will be generated. With the increase of the value of the parameter, the number of clusters is effectively reduced.

[0038] As can be seen from the above, a cascade defect cluster classification method based on material numerical calculation big data comprises the following specific steps:

[0039] S1: Obtain three-dimensional coordinates: obtain the three-dimensional coordinates of each point defect in the cluster;

[0040] S2: Calculate the distance and angle: calculate the distance and angle between adjacent point defects;

[0041] S3: Count the number: map the calculated distance and angle results to different intervals, and count the number of each interval;

[0042] S4: Calculate the defect cluster feature vector: describe the cluster geometry based on the distance and angle between adjacent point defects in the cluster, and calculate the feature vector of each cluster using the three-dimensional coordinates of each point defect;

[0043] S5: Dimension reduction: the obtained high-dimensional feature vector is reduced by using the UMAP algorithm to obtain the low-dimensional structure between data;

[0044] S6: Clustering: the clustering algorithm HDBSCAN based on density is used to visualize the cluster classification result, and the silhouette coefficient is used as an evaluation index of the clustering effect;

[0045] S7: Visualize the cluster classification result: process the cluster classification result to make it visual. The cascade defect cluster classification method based on material numerical calculation big data provided by the present application has the technical effect of solving the problem of large noise, low flexibility, and the influence of cluster information feature description on classification effect. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A whole flowchart of the cascade defect cluster classification method based on material numerical calculation big data is provided.

[0047] Figure 2 A flowchart of calculating the defect cluster feature vector of the cascade defect cluster classification method based on material numerical calculation big data is provided.

[0048] Figure 3 This is a dimensionality reduction flowchart of a cascade defect cluster classification method based on material numerical calculation big data proposed in the present invention.

[0049] Figure 4 This is a clustering flow chart of the cascade defect cluster classification method based on material numerical calculation big data proposed in the present invention.

[0050] Figure 5 This is a schematic diagram of the cluster structure of a cascade defect cluster classification method based on material numerical calculation big data proposed by the present invention.

[0051] Figure 6 This is a schematic diagram of feature extraction of a cluster classification method for material cascade collision numerical calculation big data based on a cascade defect cluster classification method based on material numerical calculation big data proposed by the present invention.

[0052] Figure 7 This is a pseudo code diagram of a cluster classification method for material cascade collision numerical calculation big data based on a cascade defect cluster classification method for material numerical calculation big data proposed by the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] The present invention discloses a cascade defect cluster classification method based on material numerical calculation big data, which is mainly used in the field of material irradiation effects and data mining scenarios.

[0055] Reference Figure 1 and Figure 5 , a cascade defect cluster classification method based on material numerical calculation big data, including the following specific steps:

[0056] S1: Get three-dimensional coordinates: Get the three-dimensional coordinates of each point defect in the cluster;

[0057] S2: Calculate distance and angle: calculate the distance and angle between adjacent point defects;

[0058] S3: Counting the number: Map the calculated distance and angle results to different intervals and count the number of each interval;

[0059] S4: Calculate defect cluster eigenvectors: Describe the cluster geometry based on the distance and angle between adjacent point defects within the cluster, and calculate the eigenvector of each cluster using the three-dimensional spatial coordinates of each point defect;

[0060] S5: Dimensionality reduction: The UMAP algorithm is used to reduce the high-dimensional feature vector obtained, and the low-dimensional structure between the data is obtained;

[0061] S6: Clustering: The HDBSCAN density-based clustering algorithm is used to visualize the cluster classification results, and the silhouette coefficient is used as an evaluation index for clustering effect;

[0062] S7: Visualize the cluster classification results: Process the cluster classification results to make them visualized;

[0063] As shown in Figure 5 , the three-dimensional coordinates of defect atoms generated after cascade collision are obtained by numerical simulation calculation, and they are divided into multiple clusters according to the union-find set algorithm, each cluster contains several point defects, the distance and angle between adjacent point defects in each cluster are calculated, the distance and angle are extracted as feature vectors to describe the spatial geometric shape of the cluster, the multi-dimensional feature vectors are processed by the nonlinear dimensionality reduction UMAP algorithm, and then the HDBSCAN clustering algorithm is used for further clustering analysis, thereby realizing the visualization of the cluster classification results.

[0064] Referring to Figure 2 and Figure 6 , in a preferred embodiment, in S4, calculating the defect cluster feature vector includes the following specific steps:

[0065] S41: Store the feature vector: Create distance_array and angle_array to store the distance feature vector and the angle feature vector respectively;

[0066] S42: Calculate the distance between adjacent defect atoms: Traverse the point defects in the cluster, calculate the distance between two adjacent defect atoms, and the angle of the adjacent defect atoms relative to the third defect atom;

[0067] S43: Divide by the maximum distance: Divide each distance by the maximum distance, and the distance range is now 0≤d≤1;

[0068] As shown in Figure 6 , two clusters with similar structures are composed of vacancies and interstitial atoms, and the significance of calculating the distance and angle based on adjacent point defects is that the corresponding ring part will have the same value, and the structure with a tail will produce additional values attached to the histogram, so the corresponding feature vectors of the two clusters have the same part, indicating that the two cluster structures are similar.

[0069] Referring to Figure 2 and Figure 7 , in a preferred embodiment, in S4, calculating the defect cluster feature vector includes the following specific steps:

[0070] S44: Processing interval distance number: divide m intervals with the maximum distance, map each distance to the corresponding interval, count the number of distances falling into each interval, and perform normalization processing to obtain an m-dimensional distance feature vector;

[0071] S45: Divide by interval angle: divide each angle by the interval angle a, where the interval angle is set according to the number of intervals n to be divided, count the number of angles falling into each interval, and perform normalization processing to obtain an n-dimensional angle feature vector;

[0072] S46: Constructing defect cluster feature vector: the m-dimensional distance feature vector and the n-dimensional angle feature vector calculated by the above steps constitute the m+n-dimensional feature vector of the defect cluster, and jump to dimension reduction;

[0073] As shown in Figure 7 , the figure is an m+n-dimensional defect cluster feature vector composed of an m-dimensional distance feature vector and an n-dimensional angle feature vector, wherein the feature vector corresponding to the distance is mapped to n intervals by the quotient of the distance between adjacent defect atoms and the maximum distance, and the feature vector corresponding to the angle is mapped to m intervals by the quotient of the angle between adjacent defect atoms and the set interval angle.

[0074] Referring to Figure 2 , in a preferred embodiment, in S42, the angle formed by the two defect atoms relative to the third defect atom in the adjacent defect atom distance is in the range of 0≤θ≤180°.

[0075] Referring to Figure 3 , in a preferred embodiment, in S5, the dimension reduction includes the following specific steps:

[0076] S51: Setting parameters: setting the parameters n_components, n_neighbors, min_dist in the UMAP algorithm to achieve different dimension reduction effects;

[0077] S52: Constructing a neighbor graph: assuming that the m+n high-dimensional feature vector data samples of the cluster are uniformly distributed in the topological space, using the KNN algorithm to find the nearest neighbor, and constructing a weighted K-neighbor graph of the defect feature vector distribution of each point in the cluster;

[0078] S53: Executing algorithm: mapping the obtained high-dimensional cluster feature vector to a low-dimensional space, executing the UMAP algorithm, and realizing dimension reduction.

[0079] Referring to Figure 3In a preferred embodiment, S51, in the parameter setting, the parameter n_components is the dimension of the reduced cluster feature space, and n_components is set to 2 to reduce the cluster feature space to two dimensions for direct visualization and to distinguish cluster types. The parameter n_neighbors is the number of nearest neighbors, and a smaller value is mainly based on the local structure of the cluster for division, and a larger value tends to the overall structure of the cluster. The min_dist is the minimum distance between each cluster high-dimensional feature vector space reduced to a low-dimensional space.

[0080] Referring to Figure 4 In a preferred embodiment, S6, the clustering includes the following specific steps:

[0081] S61: Set the parameter clustering effect: Set the parameters min_cluster_size and min_samples in the HDBSCA algorithm;

[0082] S62: Execute the HDBSCAN algorithm: Execute the HDBSCAN algorithm according to the low-dimensional space feature vectors obtained after dimension reduction to perform clustering and obtain the cluster class labels after classification;

[0083] S63: Calculate the clustering result: Take the low-dimensional space feature vectors and labels as input to calculate the clustering result, and take the silhouette coefficient as the evaluation index of cluster clustering. The value range of the silhouette coefficient is [-1, 1], the closer the distance between cluster samples with similar geometric shapes, the farther the distance between cluster samples with different geometric shapes, and the higher the value;

[0084] S64: Determine the silhouette coefficient value: If the silhouette coefficient value is not ideal, jump to S5, the dimension reduction step, and re-set the parameters until the clustering effect reaches the ideal effect.

[0085] Referring to Figure 4 and Figure 7 In a preferred embodiment, in S61, in the parameter setting for clustering effect, the parameter min_cluster_size is the number of clusters contained in each cluster class, and the larger the parameter value, the fewer the cluster classes. The parameter min_samples is the number of clusters contained in the neighborhood of a point, and the larger the parameter value, the more conservative the clustering, and more points are considered as noise.

[0086] As shown in Figure 7 , the pseudo code for feature extraction of clusters is shown in the figure, wherein the feature vectors are calculated based on the distance and angle of adjacent point defects.

[0087] Working principle: the cluster geometry is described based on the distance and angle of adjacent point defects in the cluster, the feature vector of each cluster is calculated by using the three-dimensional spatial coordinates of each point defect, the high-dimensional feature vector obtained is reduced by using the UMAP algorithm, the low-dimensional structure between the data is obtained, thereby the material irradiation cascade collision defect clusters are identified and classified, then the material defect cluster evolution mechanism is mined, the relationship between the experimental environment and the characteristics and properties of the cluster is established, the cluster classification method can simply and flexibly represent the cluster geometry shape and has high calculation efficiency, the noise caused by the same geometric shape cluster under the influence of translation, scaling and rotation is eliminated, the global and local relationship between the clusters is qualitatively determined, the relationship between the experimental environment and the characteristics and properties of the cluster is analyzed, which lays a foundation for mining and discovering the properties of the cluster and its influence, and solves the problem that the cluster information characteristic description has a large noise and low flexibility, thereby affecting the classification effect.

[0088] The above merely describes a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for classifying cascading defect clusters based on material numerical calculation big data, characterized in that, The method comprises the following specific steps: S1: obtaining three-dimensional coordinates: obtaining three-dimensional coordinates of each point defect in the cluster; S2: calculating distance and angle: calculating the distance and angle between adjacent point defects; S3: counting the number: mapping the calculated distance and angle results to different intervals, and counting the number of each interval; S4: calculating the defect cluster feature vector: describing the cluster geometry based on the distance and angle between adjacent point defects in the cluster, and calculating the feature vector of each cluster using the three-dimensional coordinates of each point defect; S5: dimension reduction: using the UMAP algorithm to reduce the high-dimensional feature vector obtained, and obtaining the low-dimensional structure between the data; S6: clustering: using the density-based clustering algorithm HDBSCAN to visualize the cluster classification results, and taking the contour coefficient as an evaluation index of the clustering effect; S7: visualizing the cluster classification results: processing the cluster classification results to make them visualized.

2. The method according to claim 1, wherein, In S4, calculating the defect cluster feature vector comprises the following specific steps: S41: storing the feature vector: creating distance_array and angle_array to store the distance feature vector and the angle feature vector respectively; S42: calculating the distance between adjacent defect atoms: traversing the point defects in the cluster, calculating the distance between two adjacent defect atoms, and the angle of the adjacent defect atoms relative to the third defect atom; S43: dividing by the maximum distance: dividing each distance by the maximum distance, and the distance range is 0≤d≤1 at this time.

3. The method of claim 2, wherein, In S4, calculating the defect cluster feature vector comprises the following specific steps: S44: processing the interval distance number: dividing m intervals according to the maximum distance, mapping each distance to the corresponding interval, counting the number of distances falling into each interval, and performing normalization processing to obtain an m-dimensional distance feature vector; S45: dividing by the interval angle: dividing each angle by the interval angle α, wherein the interval angle is set according to the number n of intervals to be divided, counting the number of angles falling into each interval, and performing normalization processing to obtain an n-dimensional angle feature vector; S46: constructing the defect cluster feature vector: the m-dimensional distance feature vector and the n-dimensional angle feature vector calculated by the above steps constitute the m+n-dimensional feature vector of the defect cluster, and jump to dimension reduction.

4. The method of claim 2, wherein, In S42, the angle range formed by two defect atoms relative to the third defect atom is 0≤θ≤180°.

5. The method of claim 1, wherein, In S5, the dimension reduction comprises the following specific steps: S51: setting parameters: setting the parameters n_components, n_neighbors, and min_dist in the UMAP algorithm to achieve different dimension reduction effects; S52: constructing a neighbor graph: assuming that the m+n high-dimensional feature vector data samples of the cluster are uniformly distributed in a topological space, using the KNN algorithm to find the nearest neighbor, and constructing a weighted K-neighbor graph of the feature vector distribution of each point defect in the cluster; S53: executing the algorithm: mapping the obtained high-dimensional cluster feature vector to a low-dimensional space, executing the UMAP algorithm, and realizing dimension reduction.

6. The method of claim 5, wherein the method is characterized by: The S51, in the parameter setting, the parameter n_components is the dimension of the cluster feature space after dimension reduction, and n_components is set to 2, so that the cluster feature space is reduced to two dimensions for direct visualization, and the cluster types are distinguished. The parameter n_neighbors is the number of nearest neighbors, and a smaller value is mainly based on the local structure of the cluster for division, and a larger value tends to the overall structure of the cluster. The min_dist is the minimum distance between each cluster high-dimensional feature vector space and the low-dimensional space.

7. The method of claim 1, wherein the method is based on material numerical calculation of big data for cascade defect cluster classification. The S6 includes the following specific steps: S61: Set the parameter clustering effect: set the parameters min_cluster_size and min_samples in the HDBSCA algorithm; S62: Execute the HDBSCAN algorithm: execute the HDBSCAN algorithm according to the low-dimensional space feature vector obtained after dimension reduction, perform clustering, and obtain the cluster class label after classification; S63: Calculate the clustering result: input the low-dimensional space feature vector and label, calculate the good and bad of the clustering result, and take the silhouette coefficient as the evaluation index of cluster clustering. The value range of the silhouette coefficient is [-1, 1]. The closer the distance between the cluster samples with similar geometric shapes, the farther the distance between the cluster samples with different geometric shapes, and the higher the value; S64: Determine the silhouette coefficient value: if the silhouette coefficient value is not ideal, jump to S5, the dimension reduction step, and reset the parameters until the clustering effect reaches the ideal effect.

8. The method of claim 7, wherein the method is characterized by: In the S61, in the parameter setting of the clustering effect, the parameter min_cluster_size is the number of clusters contained in each cluster category, and the larger the parameter value, the fewer the cluster categories. The parameter min_samples is the number of clusters contained in the neighborhood of a point, and the larger the parameter value, the more conservative the clustering, and more points are considered as noise.

Images