123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580 |
- # Import necessary libraries
- import pandas as pd
- import numpy as np
- import matplotlib.pyplot as plt
- from sklearn.cluster import KMeans
- # Import scalers and metrics from scikit-learn
- from sklearn.preprocessing import StandardScaler, RobustScaler, LabelEncoder # RobustScaler is used for scaling
- from sklearn.metrics import silhouette_score, precision_score, recall_score, f1_score, roc_auc_score, roc_curve, confusion_matrix, classification_report
- import argparse # For parsing command line arguments
- import os # For path manipulation
- import seaborn as sns # For enhanced data visualization (like confusion matrix)
-
- # Command line arguments setup
- # Defines the command line interface for the script, allowing users to specify parameters
- parser = argparse.ArgumentParser(description='Anomaly detection using K-Means clustering with visualization.')
- # --timesteps: Number of data points in each sequence (time window)
- parser.add_argument('--timesteps', type=int, default=20, help='Number of timesteps for sequences.')
- # --n_clusters: Number of clusters for K-Means. Expected to be number of failure types + normal state.
- parser.add_argument('--n_clusters', type=int, default=5, help='Number of clusters for K-Means (should match the number of failure types + normal).')
- # --n_init: Number of times K-Means is run with different initial centroids. The best result is chosen.
- parser.add_argument('--n_init', type=int, default=10, help='Number of initializations for K-Means.')
- # --transition: Flag to use data files that include transition periods for testing.
- parser.add_argument('--transition', action='store_true', help='Use transition data for testing.')
- # Plotting flags: control which plots are displayed.
- parser.add_argument('--plot_raw', action='store_true', help='Plot raw data.')
- parser.add_argument('--plot_clustered', action='store_true', help='Plot clustered data.')
- parser.add_argument('--plot_anomalies', action='store_true', help='Plot detected anomalies (based on clusters).')
- parser.add_argument('--plot_misclassified', action='store_true', help='Plot misclassified instances (based on clusters).')
- # Parse the arguments provided by the user
- options = parser.parse_args()
-
- # Assign parsed arguments to variables
- n_clusters = options.n_clusters
- timesteps = options.timesteps
- n_init = options.n_init
-
- #####################################################################################################
- # Data File Configuration
- #####################################################################################################
-
- # Number of distinct failure types we have data for (excluding the normal state)
- NumberOfFailures = 4 # So far, we have only data for the first 4 types of failures
- # List to hold file paths for training and testing data
- # datafiles[0]: training data files, datafiles[1]: testing data files
- # Inner lists correspond to different classes/failure types (0: Normal, 1-4: Failure Types)
- datafiles = [[], []] # 0 for train, 1 for test
- # Initialize inner lists for each class (Normal + NumberOfFailures)
- for i in range(NumberOfFailures + 1):
- datafiles[0].append([])
- datafiles[1].append([])
-
- # Assign specific filenames to each class for the training set
- # datafiles[0][0]: Normal training data
- # datafiles[0][1]: Failure Type 1 training data
- # ... and so on
- datafiles[0][0] = ['2024-08-07_5_', '2024-08-08_5_', '2025-01-25_5_', '2025-01-26_5_']
- datafiles[0][1] = ['2024-12-11_5_', '2024-12-12_5_', '2024-12-13_5_']
- datafiles[0][2] = ['2024-12-18_5_', '2024-12-21_5_', '2024-12-22_5_', '2024-12-23_5_', '2024-12-24_5_']
- datafiles[0][3] = ['2024-12-28_5_', '2024-12-29_5_', '2024-12-30_5_']
- datafiles[0][4] = ['2025-02-13_5_', '2025-02-14_5_']
-
- # Assign specific filenames for the testing set
- # Uses different files based on whether the --transition flag is set
- if options.transition:
- # Test files including transition data
- datafiles[1][0] = ['2025-01-27_5_', '2025-01-28_5_']
- datafiles[1][1] = ['2024-12-14_5_', '2024-12-15_5_', '2024-12-16_5_'] # with TRANSITION
- datafiles[1][2] = ['2024-12-17_5_', '2024-12-19_5_', '2024-12-25_5_', '2024-12-26_5_'] # with TRANSITION
- datafiles[1][3] = ['2024-12-27_5_', '2024-12-31_5_', '2025-01-01_5_'] # with TRANSITION
- datafiles[1][4] = ['2025-02-12_5_', '2025-02-15_5_', '2025-02-16_5_']
- else:
- # Test files without explicit transition data
- datafiles[1][0] = ['2025-01-27_5_', '2025-01-28_5_']
- datafiles[1][1] = ['2024-12-14_5_', '2024-12-15_5_']
- datafiles[1][2] = ['2024-12-19_5_', '2024-12-25_5_', '2024-12-26_5_']
- datafiles[1][3] = ['2024-12-31_5_', '2025-01-01_5_']
- datafiles[1][4] = ['2025-02-15_5_', '2025-02-16_5_']
-
- # Features (columns) to be used from the data files
- features = ['r1 s1', 'r1 s4', 'r1 s5']
- # Store the initial count of features before potentially adding derived features
- n_original_features = len(features) # Store the original number of features
-
- # Dictionaries to map feature names to display names (e.g., for plots)
- featureNames = {}
- featureNames['r1 s1'] = r'<span class="math-inline">T\_\{evap\}</span>' # Evaporator Temperature
- featureNames['r1 s4'] = r'<span class="math-inline">T\_\{cond\}</span>' # Condenser Temperature
- featureNames['r1 s5'] = r'<span class="math-inline">T\_\{air\}</span>' # Air Temperature
-
- # Dictionaries to map feature names to their units (e.g., for plots)
- unitNames = {}
- unitNames['r1 s1'] = r'($^o$C)'
- unitNames['r1 s4'] = r'($^o$C)'
- unitNames['r1 s5'] = r'($^o$C)'
-
- # Redundant variable, but kept from original code
- NumFeatures = len(features)
-
- #####################################################################################################
- # Data Loading and Preprocessing (Training Data)
- #####################################################################################################
-
- # List to hold DataFrames for training data, organized by class
- dataTrain = []
- # Loop through each list of files for each training class
- for class_files in datafiles[0]:
- class_dfs = [] # List to hold dataframes for current class
- # Loop through each filename in the current class
- for base_filename in class_files:
- # Construct the full file path
- script_dir = os.path.dirname(os.path.abspath(__file__)) # Get directory of the current script
- data_dir = os.path.join(script_dir, 'data') # Assume data is in a 'data' subdirectory
- filepath = os.path.join(data_dir, f'{base_filename}.csv') # Full path to the CSV file
- try:
- # Read the CSV file into a pandas DataFrame
- df = pd.read_csv(filepath)
- # Convert 'datetime' column to datetime objects using two possible formats, coercing errors
- df['timestamp'] = pd.to_datetime(df['datetime'], format='%m/%d/%Y %H:%M', errors='coerce')
- df['timestamp'] = df['timestamp'].fillna(pd.to_datetime(df['datetime'], format='%d-%m-%Y %H:%M:%S', errors='coerce'))
- # Convert feature columns to numeric, coercing errors to NaN
- for col in features:
- df[col] = pd.to_numeric(df[col], errors='coerce')
- # Set the timestamp as index, resample to 5-minute frequency, and calculate the mean for features
- df = df.set_index('timestamp').resample('5Min')[features].mean() # Resample and calculate mean only for features
- # Estimate missing values (NaN) using linear interpolation
- df = df[features].interpolate() # Estimate missing values using linear interpolation
- # Append the processed DataFrame to the list for the current class
- class_dfs.append(df)
- except FileNotFoundError:
- # Print a warning if a file is not found and skip it
- print(f"Warning: File {filepath} not found and skipped.")
- # If any files were successfully loaded for this class, concatenate them
- if class_dfs:
- dataTrain.append(pd.concat(class_dfs))
-
- # Concatenate all class DataFrames into a single DataFrame for training
- combined_train_data = pd.concat(dataTrain)
-
- #####################################################################################################
- # Data Loading and Preprocessing (Test Data)
- #####################################################################################################
-
- # List to hold DataFrames for test data, organized by class
- # Each element in dataTest corresponds to a different class (Normal, Failure Type 1, etc.)
- dataTest = []
- # Loop through each list of files for each test class
- for class_files in datafiles[1]:
- class_dfs = [] # List to hold dataframes for current class
- # Loop through each filename in the current class
- for base_filename in class_files:
- # Construct the full file path
- script_dir = os.path.dirname(os.path.abspath(__file__)) # Get directory of the current script
- data_dir = os.path.join(script_dir, 'data') # Assume data is in a 'data' subdirectory
- filepath = os.path.join(data_dir, f'{base_filename}.csv') # Full path to the CSV file
- try:
- # Read the CSV file into a pandas DataFrame
- df = pd.read_csv(filepath)
- # Convert 'datetime' column to datetime objects using two possible formats, coercing errors
- df['timestamp'] = pd.to_datetime(df['datetime'], format='%m/%d/%Y %H:%M', errors='coerce')
- df['timestamp'] = df['timestamp'].fillna(pd.to_datetime(df['datetime'], format='%d-%m-%Y %H:%M:%S', errors='coerce'))
- # Convert feature columns to numeric, coercing errors to NaN
- for col in features:
- df[col] = pd.to_numeric(df[col], errors='coerce')
- # Set the timestamp as index, resample to 5-minute frequency, and calculate the mean for features
- df = df.set_index('timestamp').resample('5Min')[features].mean() # Resample and calculate mean only for features
- # Estimate missing values (NaN) using linear interpolation
- df = df[features].interpolate() # Estimate missing values using linear interpolation
- # Append the processed DataFrame to the list for the current class
- class_dfs.append(df)
- except FileNotFoundError:
- # Print a warning if a file is not found and skip it
- print(f"Warning: File {filepath} not found and skipped.")
- # If any files were successfully loaded for this class, concatenate them
- if class_dfs:
- dataTest.append(pd.concat(class_dfs))
-
- #####################################################################################################
- # Raw Data Plotting (Optional)
- #####################################################################################################
-
- # Plot raw data if the --plot_raw flag is provided
- if options.plot_raw:
- num_features = len(features)
- # Create a figure and a set of subplots (one for each feature)
- fig, axes = plt.subplots(num_features, 1, figsize=(15, 5 * num_features), sharex=True)
- # Ensure axes is an array even if there's only one feature
- if num_features == 1:
- axes = [axes]
- # Loop through each feature
- for i, feature in enumerate(features):
- # Loop through each test data DataFrame (each class)
- for k, df in enumerate(dataTest):
- # Plot the feature data over time for the current class
- axes[i].plot(df.index, df[feature], label=f'Class {k}')
- # Set ylabel and title for the subplot
- axes[i].set_ylabel(f'{featureNames[feature]} {unitNames[feature]}')
- axes[i].set_title(featureNames[feature])
- # Add legend to the subplot
- axes[i].legend()
- # Adjust layout to prevent labels overlapping
- plt.tight_layout()
- # Display the plot
- plt.show()
- # exit(0) # Uncomment to exit after plotting raw data
-
- ########################################################################################################
- # Data Scaling
- ########################################################################################################
-
- # Initialize the scaler (RobustScaler is less affected by outliers than StandardScaler)
- # StandardScaler() # Original scaler
- scaler = RobustScaler() # Changed from StandardScaler
-
- # Fit the scaler on the training data and transform it
- # Only the original features are scaled
- scaled_train_data = scaler.fit_transform(combined_train_data[features]) # Normalize only the original features
-
- # Transform the test data using the scaler fitted on the training data
- # A list comprehension is used to transform each test DataFrame
- scaled_test_data_list = [scaler.transform(df[features]) for df in dataTest] # Normalize only the original features
-
- # Convert normalized data back to pandas DataFrames for easier handling (optional but can be useful)
- scaled_train_df = pd.DataFrame(scaled_train_data, columns=features, index=combined_train_data.index)
- scaled_test_df_list = [pd.DataFrame(data, columns=features, index=df.index) for data, df in zip(scaled_test_data_list, dataTest)]
-
- ############################################################################################################
- # Sequence Creation with Rate of Change Feature Engineering
- ############################################################################################################
-
- # Function to create time sequences from data and append the rate of change as new features
- def create_sequences_with_rate_of_change(data, timesteps, original_features_count): # Parameter name indicates count
- sequences = [] # List to store the created sequences
- # Iterate through the data to create overlapping sequences
- for i in range(len(data) - timesteps + 1):
- # Extract a sequence of 'timesteps' length
- sequence = data[i:i + timesteps]
- # Calculate the difference between consecutive points along the time axis (axis=0)
- # This computes the rate of change for each feature across timesteps
- rate_of_change = np.diff(sequence[:timesteps], axis=0)
- # Pad the rate of change to have the same number of timesteps as the original sequence
- # np.diff reduces the number of timesteps by 1, so we add a row of zeros at the beginning
- # Use the count of original features for padding dimension
- padding = np.zeros((1, original_features_count)) # Corrected: Use the features count
- rate_of_change_padded = np.vstack((padding, rate_of_change)) # Stack the padding on top
- # Concatenate the original sequence and the padded rate of change sequence horizontally
- # Resulting sequence has 'timesteps' rows and '2 * original_features_count' columns
- sequences.append(np.hstack((sequence, rate_of_change_padded))) # Concatenate original and rate of change
- # Convert the list of sequences into a NumPy array
- return np.array(sequences)
-
- # Create time sequences with rate of change for the scaled training data
- # The output shape will be (num_training_sequences, timesteps, 2 * n_original_features)
- X_train_sequences = create_sequences_with_rate_of_change(scaled_train_df.values, timesteps, n_original_features) # Pass n_original_features
-
- # Create time sequences with rate of change for each scaled test data DataFrame
- # X_test_sequences_list will be a list of arrays, one for each test class
- X_test_sequences_list = [create_sequences_with_rate_of_change(df.values, timesteps, n_original_features) for df in scaled_test_df_list] # Pass n_original_features
-
- ############################################################################################################
- # K-Means Clustering Model
- ############################################################################################################
-
- # Reshape the training sequences for K-Means
- # K-Means expects a 2D array (samples, features)
- # We flatten each sequence (timesteps * total_features) into a single row
- n_samples, n_timesteps, n_total_features = X_train_sequences.shape
- X_train_reshaped = X_train_sequences.reshape(n_samples, n_timesteps * n_total_features)
-
- # Train the K-Means model
- # n_clusters: Number of clusters (expected to be number of classes)
- # random_state=42: Ensures reproducibility of initial centroids for n_init runs
- # n_init=10: Runs K-Means 10 times with different centroid seeds and picks the best result (lowest inertia)
- kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init=n_init) # n_init to avoid convergence to local optima
- # Fit the K-Means model on the reshaped training data
- kmeans.fit(X_train_reshaped)
-
- ############################################################################################################################
- # Predict Clusters for Test Data
- ############################################################################################################################
-
- # List to store predicted cluster labels for each test data DataFrame
- cluster_labels_test_list = []
- # List to store reshaped test data (useful for evaluation metrics later)
- X_test_reshaped_list = []
- # kmeans_models = [] # To store kmeans model for each test set (This variable is declared but not used subsequently)
-
- # Loop through each test data sequence array (each class)
- for i, X_test_seq in enumerate(X_test_sequences_list):
- # Get dimensions of the current test sequence array
- n_samples_test, n_timesteps_test, n_total_features_test = X_test_seq.shape
- # Reshape the test sequences for prediction (flatten each sequence)
- X_test_reshaped = X_test_seq.reshape(n_samples_test, n_timesteps_test * n_total_features_test)
- # Predict cluster labels for the reshaped test data
- labels = kmeans.predict(X_test_reshaped)
- # Append the predicted labels and reshaped data to the lists
- cluster_labels_test_list.append(labels)
- X_test_reshaped_list.append(X_test_reshaped) # Append reshaped data
- # kmeans_models.append(kmeans) # Store the trained kmeans model (Variable declared but not used)
-
- ############################################################################################################################
- # Plotting Clustered Data (Optional)
- ############################################################################################################
-
- # Function to plot the original data points colored by their assigned cluster label
- # Plots only the original features
- def plot_clustered_data(original_data_list, cluster_labels_list, n_clusters, features, featureNames, unitNames):
- num_features = len(features)
- # Create subplots, one for each original feature
- fig, axes = plt.subplots(num_features, 1, figsize=(15, 5 * num_features), sharex=True)
- # Ensure axes is an array even if only one feature
- if num_features == 1:
- axes = [axes]
- # Generate a color map for the clusters
- colors = plt.cm.viridis(np.linspace(0, 1, n_clusters)) # Assign colors to each cluster
-
- # Loop through each original test data DataFrame (each class)
- for k, df in enumerate(original_data_list):
- original_indices = df.index # Get the original time index
- # Get the time index corresponding to the start of each sequence (shifted by timesteps-1)
- time_index = original_indices[timesteps - 1:]
-
- # Loop through each original feature
- for i, feature in enumerate(features):
- # Loop through each predicted cluster ID
- for cluster_id in range(n_clusters):
- # Find the indices in the current test data corresponding to the current cluster ID
- cluster_indices_kmeans = np.where(cluster_labels_list[k] == cluster_id)[0]
- # If there are data points assigned to this cluster
- if len(cluster_indices_kmeans) > 0:
- # Scatter plot the data points for this cluster
- # x-axis: time_index points corresponding to the sequence end
- # y-axis: original feature values at those time_index points
- # color: color assigned to the cluster
- # label: label for the cluster (only show for the first class (k==0) to avoid redundant legends)
- # s=10: size of the scatter points
- axes[i].scatter(time_index[cluster_indices_kmeans], df[feature].loc[time_index[cluster_indices_kmeans]], color=colors[cluster_id], label=f'Cluster {cluster_id}' if k == 0 else "", s=10)
- # Set ylabel and title for the subplot
- axes[i].set_ylabel(f'{featureNames[feature]} {unitNames[feature]}')
- axes[i].set_title(featureNames[feature])
- # Add legend to the last subplot (or each if desired)
- axes[num_features - 1].legend(loc='upper right') # Place legend on the last subplot
-
- # Adjust layout and display the plot
- plt.tight_layout()
- plt.show()
-
- # Call the plotting function if the --plot_clustered flag is provided
- if options.plot_clustered:
- plot_clustered_data(dataTest, cluster_labels_test_list, n_clusters, features, featureNames, unitNames)
-
- #####################################################################################################
- # Evaluation and plotting of anomalies and misclassified instances (based on cluster labels)
- #####################################################################################################
-
- # Function to evaluate clustering results and plot anomalies/misclassified instances
- def evaluate_and_plot_anomalies(kmeans_model, scaled_test_data_list, n_clusters, original_test_data_list, true_labels_list, features, featureNames, unitNames, plot_anomalies=False, plot_misclassified=False):
- # Lists to store collected data and labels across all test classes
- all_y_true_categorical = [] # Stores true labels (0, 1, 2, ...) for each sequence
- all_predicted_cluster_labels = [] # Stores predicted cluster ID for each sequence
- all_original_test_sequences = [] # Stores the original feature values for each sequence (for plotting)
-
- # Lists to store evaluation metrics per test class (before combining)
- inertia_values = [] # Inertia values for each class's data predicted by the model
- silhouette_scores = [] # Silhouette scores for each class's data predicted by the model
-
- # Loop through each test class data (scaled, original, and true labels)
- for i, (scaled_test_df, original_test_df, y_true_categorical) in enumerate(zip(scaled_test_data_list, original_test_data_list, true_labels_list)):
- # Create sequences with rate of change for the current scaled test data
- X_test_sequences = create_sequences_with_rate_of_change(scaled_test_df.values, timesteps, n_original_features) # Pass n_original_features
- # Skip evaluation for this class if no sequences were generated (data too short)
- if X_test_sequences.size == 0:
- print(f"Warning: No test sequences generated for class {i}. Skipping evaluation for this class.")
- continue
- # Reshape the sequences for prediction by the trained K-Means model
- n_samples_test = X_test_sequences.shape[0]
- X_test_reshaped = X_test_sequences.reshape(n_samples_test, -1)
- # Predict cluster labels for the current test class data
- cluster_labels_predicted = kmeans_model.predict(X_test_reshaped)
-
- # Calculate and store Inertia for the current class's data (based on the overall model)
- # This is different from the model's final inertia on training data
- inertia_values.append(kmeans_model.inertia_) # Note: This seems to append the total model inertia, not per-class inertia. It might be intended to be calculated differently here. Keeping original code logic.
- # Calculate and store Silhouette score if possible (requires >1 unique labels and >0 samples)
- if len(np.unique(cluster_labels_predicted)) > 1 and len(cluster_labels_predicted) > 0:
- silhouette_scores.append(silhouette_score(X_test_reshaped, cluster_labels_predicted))
- else:
- silhouette_scores.append(np.nan) # Append NaN if silhouette cannot be calculated
-
- # Get the time indices corresponding to the end of each sequence in the original data
- original_indices = original_test_df.index[timesteps - 1:]
-
- # Collect true labels, predicted labels, and original sequences for evaluation/plotting
- # Loop through the sequences generated for the current class
- for j, label in enumerate(y_true_categorical[timesteps - 1:]): # Iterate over true labels corresponding to sequence ends
- all_y_true_categorical.append(label) # Append the true label
- all_predicted_cluster_labels.append(cluster_labels_predicted[j]) # Append the predicted cluster label
- # Get the start and end index in the original DataFrame for the current sequence
- start_index = original_test_df.index.get_loc(original_indices[j]) - (timesteps - 1)
- end_index = start_index + timesteps
- # Extract and append the original feature values for the current sequence
- all_original_test_sequences.append(original_test_df[features].iloc[start_index:end_index].values) # Append
-
- # Convert collected lists to NumPy arrays for easier handling
- all_y_true_categorical = np.array(all_y_true_categorical)
- all_predicted_cluster_labels = np.array(all_predicted_cluster_labels)
- all_original_test_sequences = np.array(all_original_test_sequences)
-
- # Print evaluation metrics (based on collected values across all test classes)
- print("\nEvaluation Metrics:")
- # Print mean Inertia (likely the final Inertia of the trained model as per the loop)
- print(f"Inertia (final): {np.mean(inertia_values):.4f}") # Check if this is the intended calculation
- # Print mean Silhouette score across classes (ignoring NaNs)
- print(f"Average Silhouette Score (valid cases): {np.nanmean(silhouette_scores):.4f}")
-
- # Analyze clusters and assign a dominant true label to each cluster ID
- # This helps in mapping cluster IDs back to meaningful class labels for evaluation
- cluster_dominant_label = {} # Dictionary to store the dominant true label for each cluster ID
- for cluster_id in range(n_clusters): # Loop through each cluster ID
- # Find indices of all sequences assigned to the current cluster ID
- indices_in_cluster = np.where(all_predicted_cluster_labels == cluster_id)[0]
- # If there are sequences in this cluster
- if len(indices_in_cluster) > 0:
- # Get the true labels for all sequences in this cluster
- labels_in_cluster = all_y_true_categorical[indices_in_cluster]
- # If there are labels (and thus samples) in this cluster
- if len(labels_in_cluster) > 0:
- # Find the most frequent true label (dominant label) in this cluster
- dominant_label = np.argmax(np.bincount(labels_in_cluster))
- cluster_dominant_label[cluster_id] = dominant_label # Store the dominant label
- else:
- cluster_dominant_label[cluster_id] = -1 # Assign -1 if no data points have true labels (shouldn't happen if indices_in_cluster > 0 and all_y_true_categorical is aligned)
- else:
- cluster_dominant_label[cluster_id] = -1 # Assign -1 if the cluster is empty
-
- # Create predicted labels in numeric form based on the dominant true label of the assigned cluster
- # This maps the predicted cluster ID for each sequence to the dominant true label of that cluster
- predicted_labels_numeric = np.array([cluster_dominant_label.get(cluster_id, -1) for cluster_id in all_predicted_cluster_labels])
-
- # Evaluate the clustering's ability to separate classes using classification metrics
- # Only consider instances where a dominant label could be assigned (predicted_labels_numeric != -1)
- valid_indices = predicted_labels_numeric != -1 # Indices where a dominant label mapping exists
- # Perform evaluation if there are valid instances and more than one true class represented
- if np.sum(valid_indices) > 0 and len(np.unique(all_y_true_categorical[valid_indices])) > 1:
- print("\nEvaluation Results (Clusters vs True Labels):")
- # Print classification report (Precision, Recall, F1-score per class, and overall metrics)
- print(classification_report(all_y_true_categorical[valid_indices], predicted_labels_numeric[valid_indices]))
- # Compute the confusion matrix
- cm = confusion_matrix(all_y_true_categorical[valid_indices], predicted_labels_numeric[valid_indices])
- # Plot the confusion matrix using seaborn heatmap
- plt.figure(figsize=(8, 6))
- sns.heatmap(cm, annot=True, fmt='d', cmap='Blues') # annot=True shows values, fmt='d' formats as integers
- plt.xlabel('Predicted Cluster (Dominant True Label)') # Label for x-axis
- plt.ylabel('True Label') # Label for y-axis
- plt.title('Confusion Matrix (Clusters vs True Labels)') # Title of the plot
- plt.show() # Display the plot
- else:
- print("\nCould not perform detailed evaluation (not enough data or classes).")
-
- #################################################################################################
- # Plotting Anomalies (Optional)
- #################################################################################################
-
- # Plot detected anomalies if the --plot_anomalies flag is provided
- # Anomalies are defined here as instances assigned to clusters whose dominant true label is > 0 (Failure types)
- if plot_anomalies:
- print("\nChecking anomaly data:")
- # Identify clusters that predominantly contain non-normal true labels (failure types)
- anomaly_clusters = [cluster_id for cluster_id, label in cluster_dominant_label.items() if label > 0]
- # Find indices of all sequences assigned to these "anomaly" clusters
- anomaly_indices = np.where(np.isin(all_predicted_cluster_labels, anomaly_clusters))[0]
- # If any anomalies are detected
- if len(anomaly_indices) > 0:
- # Limit the number of anomaly plots to show
- num_anomalies_to_plot = min(5, len(anomaly_indices))
- colors = ['red', 'green', 'blue'] # Define different colors for features
- # Randomly select and plot a few anomaly sequences
- for i in np.random.choice(anomaly_indices, num_anomalies_to_plot, replace=False):
- # Print shape and sample values for the sequence being plotted
- print(f"Shape of all_original_test_sequences[{i}]: {all_original_test_sequences[i].shape}")
- print(f"First few values of all_original_test_sequences[{i}]:\n{all_original_test_sequences[i][:5]}")
- # Create a new figure for each anomaly plot
- plt.figure(figsize=(12, 6))
- # Plot each feature in the sequence over time steps
- for j, feature in enumerate(features):
- # Plot the feature values (y-axis) against time steps (x-axis)
- plt.plot(np.arange(timesteps), all_original_test_sequences[i][:, j], label=feature, color=colors[j % len(colors)])
- # Get the true label and predicted cluster for the title
- true_label = all_y_true_categorical[i]
- predicted_cluster_for_title = all_predicted_cluster_labels[i]
- # Set the title for the anomaly plot, including true label and predicted cluster
- plt.title(f'Detected Anomaly (True: {true_label}, Cluster: {predicted_cluster_for_title})') # Corrected title format
- plt.xlabel('Time Step')
- plt.ylabel('Value')
- plt.legend() # Add legend to identify features
- plt.show() # Display the plot
- else:
- print("No anomalies detected based on cluster dominance.")
-
- #################################################################################################
- # Plotting Misclassified Instances (Optional)
- #################################################################################################
-
- # Plot misclassified instances if the --plot_misclassified flag is provided
- # Misclassified are defined here as instances where the true label is DIFFERENT from the dominant label of the assigned cluster
- if plot_misclassified:
- print("\nChecking misclassified data:")
- # Find indices where the true label does not match the dominant label of the predicted cluster
- misclassified_indices = np.where(all_y_true_categorical != predicted_labels_numeric)[0]
- # If any misclassified instances are found
- if len(misclassified_indices) > 0:
- # Limit the number of misclassified plots to show
- num_misclassified_to_plot = min(5, len(misclassified_indices))
- colors = ['red', 'green', 'blue'] # Define different colors for features
- # Randomly select and plot a few misclassified sequences
- for i in np.random.choice(misclassified_indices, num_misclassified_to_plot, replace=False):
- # Print shape and sample values for the sequence being plotted
- print(f"Shape of all_original_test_sequences[{i}]: {all_original_test_sequences[i].shape}")
- print(f"First few values of all_original_test_sequences[{i}]:\n{all_original_test_sequences[i][:5]}")
- # Create a new figure for each misclassified plot
- plt.figure(figsize=(12, 6))
- # Plot each feature in the sequence over time steps
- for j, feature in enumerate(features):
- # Plot the feature values (y-axis) against time steps (x-axis)
- plt.plot(np.arange(timesteps), all_original_test_sequences[i][:, j], label=feature, color=colors[j % len(colors)])
- # FIXED: Get labels using index i for plot title
- true_label = all_y_true_categorical[i] # Get the true label
- predicted_label = predicted_labels_numeric[i] # Get the numeric predicted label (dominant cluster label)
- # Set the title for the misclassified plot, including true label and predicted cluster's dominant label
- plt.title(f'Misclassified Instance (True: {true_label}, Predicted Cluster Dominant Label: {predicted_label})') # Corrected title format
- plt.xlabel('Time Step')
- plt.ylabel('Value')
- plt.legend() # Add legend to identify features
- plt.show() # Display the plot
- else:
- print("No misclassified instances found based on cluster dominance.")
-
- # Return the true and predicted labels for potential further use
- return all_y_true_categorical, predicted_labels_numeric
-
- #####################################################################################################
- # Main Execution
- #####################################################################################################
-
- # Create the list of true labels for the test data
- # Assign a numeric label (0, 1, 2, ...) to each sequence based on its original file class
- true_labels_list = []
- for i, df in enumerate(dataTest): # Loop through each test DataFrame (each class)
- # Create a numpy array of the same length as the DataFrame, filled with the class index (i)
- true_labels_list.append(np.full(len(df), i))
-
- # Call the evaluation and plotting function with the necessary data and options
- y_true_final, y_pred_final = evaluate_and_plot_anomalies(kmeans, scaled_test_df_list, n_clusters, dataTest, true_labels_list, features, featureNames, unitNames, plot_anomalies=options.plot_anomalies, plot_misclassified=options.plot_misclassified)
-
- #####################################################################################################
- # Final Evaluation Metrics (on combined test data)
- #####################################################################################################
-
- # Calculate and print final Inertia and Silhouette Score for the combined test data
- # Check if there's any reshaped test data available
- if X_test_reshaped_list:
- # Vertically stack all reshaped test data arrays into a single array
- X_test_combined_reshaped = np.vstack(X_test_reshaped_list)
- # Concatenate all predicted cluster labels into a single array
- all_cluster_labels_test = np.concatenate(cluster_labels_test_list)
-
- # Print K-Means evaluation metrics on the combined test data
- print("\nK-Means Model Evaluation on Combined Test Data:")
- # Print the final Inertia of the trained K-Means model
- print(f"Inertia: {kmeans.inertia_:.4f}")
-
- # Calculate and print Silhouette Score if possible
- # Requires more than one unique predicted label and at least one sample
- if len(np.unique(all_cluster_labels_test)) > 1 and len(all_cluster_labels_test) > 0:
- silhouette = silhouette_score(X_test_combined_reshaped, all_cluster_labels_test)
- print(f"Silhouette Score: {silhouette:.4f}")
- else:
- print("Silhouette Score: Not applicable for single cluster.")
- else:
- # Print a message if no test data sequences were available for evaluation
- print("\nNo test data sequences available to evaluate Inertia and Silhouette Score.")
|