Enhances similarity computation by normalizing feature values and adding feature range definitions

gedi/utils/io_helpers.py

@@ -84,27 +84,140 @@ def dump_features_json(features: dict, output_path, content_type="features"):
         json.dump(features, fp, default=int)
         print(f"SUCCESS: Saved {len(features)-1} {content_type} in {json_path}")#-1 because 'log' is not a feature
 
-def compute_similarity(v1, v2):
+def normalize_value(value, min_val, max_val):
+    return (value - min_val) / (max_val - min_val) if max_val != min_val else 0.0
 
-    # Convert all values to float except for the value for the key "Log"
+def compute_similarity(v1, v2):
+    feature_ranges = bpic_feature_values()
+    
+    # Convert all values to float except for the "log" key
     v1 = {k: (float(v) if k != "log" else v) for k, v in v1.items()}
     v2 = {k: (float(v) if k != "log" else v) for k, v in v2.items()}
-
-    # Filter out non-numeric values and ensure the same keys exist in both dictionaries
-    common_keys = set(v1.keys()).intersection(set(v2.keys()))
+    
+    # Identify common numeric keys
+    common_keys = set(v1.keys()).intersection(set(v2.keys()), set(feature_ranges.keys()))
     numeric_keys = [k for k in common_keys if isinstance(v1[k], (int, float)) and isinstance(v2[k], (int, float))]
-
-    # Create vectors from the filtered keys
-    vec1 = np.array([v1[k] for k in numeric_keys])
-    vec2 = np.array([v2[k] for k in numeric_keys])
-
-    if len(vec1) == 0 or len(vec2) == 0:
-        print("[ERROR]: No common numeric keys found for (Edit) Distance calculation.")
+    
+    if not numeric_keys:
+        print("[ERROR]: No common numeric keys found for similarity calculation.")
         return None
-
-    else:
-        # Calculate Euclidean Similarity
-        target_similarity = 1 / (1 + euclidean(vec1, vec2))
-        # print("VECTORS: ", vec1, vec2, target_similarity)
-
-        return target_similarity
+    
+    # Normalize values and compute differences
+    differences = []
+    for key in numeric_keys:
+        min_val, max_val = feature_ranges[key]
+        norm_v1 = normalize_value(v1[key], min_val, max_val)
+        norm_v2 = normalize_value(v2[key], min_val, max_val)
+        differences.append(abs(norm_v1 - norm_v2))
+    
+    # Compute average difference as similarity metric
+    target_similarity = 1 - np.mean(differences)
+    return target_similarity
+
+    
+def bpic_feature_values():
+    
+    data_dict = {
+        "n_traces": [226.0, 251734.0],
+        "n_variants": [6.0, 28457.0],
+        "ratio_variants_per_number_of_traces": [0.0, 1.0],
+        "trace_len_min": [1.0, 24.0],
+        "trace_len_max": [1.0, 2973.0],
+        "trace_len_mean": [1.0, 131.49],
+        "trace_len_median": [1.0, 55.0],
+        "trace_len_mode": [1.0, 61.0],
+        "trace_len_std": [0.0, 202.53],
+        "trace_len_variance": [0.0, 41017.89],
+        "trace_len_q1": [1.0, 44.0],
+        "trace_len_q3": [1.0, 169.0],
+        "trace_len_iqr": [0.0, 161.0],
+        "trace_len_geometric_mean": [1.0, 53.78],
+        "trace_len_geometric_std": [1.0, 5.65],
+        "trace_len_harmonic_mean": [1.0, 51.65],
+        "trace_len_skewness": [-0.58, 111.97],
+        "trace_len_kurtosis": [-0.97, 14006.75],
+        "trace_len_coefficient_variation": [0.0, 4.74],
+        "trace_len_entropy": [5.33, 12.04],
+        "trace_len_hist1": [0.0, 1.99],
+        "trace_len_hist2": [0.0, 0.42],
+        "trace_len_hist3": [0.0, 0.4],
+        "trace_len_hist4": [0.0, 0.19],
+        "trace_len_hist5": [0.0, 0.14],
+        "trace_len_hist6": [0.0, 10.0],
+        "trace_len_hist7": [0.0, 0.02],
+        "trace_len_hist8": [0.0, 0.04],
+        "trace_len_hist9": [0.0, 0.0],
+        "trace_len_hist10": [0.0, 2.7],
+        "trace_len_skewness_hist": [-0.58, 111.97],
+        "trace_len_kurtosis_hist": [-0.97, 14006.75],
+        "ratio_most_common_variant": [0.0, 0.79],
+        "ratio_top_1_variants": [0.0, 0.87],
+        "ratio_top_5_variants": [0.0, 0.98],
+        "ratio_top_10_variants": [0.0, 0.99],
+        "ratio_top_20_variants": [0.2, 1.0],
+        "ratio_top_50_variants": [0.5, 1.0],
+        "ratio_top_75_variants": [0.75, 1.0],
+        "mean_variant_occurrence": [1.0, 24500.67],
+        "std_variant_occurrence": [0.04, 42344.04],
+        "skewness_variant_occurrence": [1.54, 64.77],
+        "kurtosis_variant_occurrence": [0.66, 5083.46],
+        "n_unique_activities": [1.0, 1152.0],
+        "activities_min": [1.0, 66058.0],
+        "activities_max": [34.0, 466141.0],
+        "activities_mean": [4.13, 66058.0],
+        "activities_median": [2.0, 66058.0],
+        "activities_std": [0.0, 120522.25],
+        "activities_variance": [0.0, 14525612122.34],
+        "activities_q1": [1.0, 66058.0],
+        "activities_q3": [4.0, 79860.0],
+        "activities_iqr": [0.0, 77290.0],
+        "activities_skewness": [-0.06, 15.21],
+        "activities_kurtosis": [-1.5, 315.84],
+        "n_unique_start_activities": [1.0, 809.0],
+        "start_activities_min": [1.0, 150370.0],
+        "start_activities_max": [27.0, 199867.0],
+        "start_activities_mean": [3.7, 150370.0],
+        "start_activities_median": [1.0, 150370.0],
+        "start_activities_std": [0.0, 65387.49],
+        "start_activities_variance": [0.0, 4275524278.19],
+        "start_activities_q1": [1.0, 150370.0],
+        "start_activities_q3": [4.0, 150370.0],
+        "start_activities_iqr": [0.0, 23387.25],
+        "start_activities_skewness": [0.0, 9.3],
+        "start_activities_kurtosis": [-2.0, 101.82],
+        "n_unique_end_activities": [1.0, 757.0],
+        "end_activities_min": [1.0, 16653.0],
+        "end_activities_max": [28.0, 181328.0],
+        "end_activities_mean": [3.53, 24500.67],
+        "end_activities_median": [1.0, 16653.0],
+        "end_activities_std": [0.0, 42344.04],
+        "end_activities_variance": [0.0, 1793017566.89],
+        "end_activities_q1": [1.0, 16653.0],
+        "end_activities_q3": [3.0, 39876.0],
+        "end_activities_iqr": [0.0, 39766.0],
+        "end_activities_skewness": [-0.7, 13.82],
+        "end_activities_kurtosis": [-2.0, 255.39],
+        "eventropy_trace": [0.0, 13.36],
+        "eventropy_prefix": [0.0, 16.77],
+        "eventropy_global_block": [0.0, 24.71],
+        "eventropy_lempel_ziv": [0.0, 685.0],
+        "eventropy_k_block_diff_1": [-328.0, 962.0],
+        "eventropy_k_block_diff_3": [0.0, 871.0],
+        "eventropy_k_block_diff_5": [0.0, 881.0],
+        "eventropy_k_block_ratio_1": [0.0, 935.0],
+        "eventropy_k_block_ratio_3": [0.0, 7.11],
+        "eventropy_k_block_ratio_5": [0.0, 7.11],
+        "eventropy_knn_3": [0.0, 8.93],
+        "eventropy_knn_5": [0.0, 648.0],
+        "eventropy_knn_7": [0.0, 618.0],
+        "epa_variant_entropy": [0.0, 11563842.15],
+        "epa_normalized_variant_entropy": [0.0, 0.9],
+        "epa_sequence_entropy": [0.0, 21146257.12],
+        "epa_normalized_sequence_entropy": [0.0, 0.76],
+        "epa_sequence_entropy_linear_forgetting": [0.0, 14140225.9],
+        "epa_normalized_sequence_entropy_linear_forgetting": [0.0, 0.42],
+        "epa_sequence_entropy_exponential_forgetting": [0.0, 15576076.83],
+        "epa_normalized_sequence_entropy_exponential_forgetting": [0.0, 0.51]
+    }
+    
+    return data_dict