PHM-Vibench / data_analysis.py
Richie-17's picture
Upload folder using huggingface_hub
508cc13 verified
import pandas as pd
import h5py
import numpy as np
# from scipy.stats import skew, kurtosis # No longer needed for features
import umap
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import seaborn as sns
import os
import concurrent.futures
from functools import partial
# --- Configuration ---
TRUNCATE_M_POINTS = 2048 # Default truncation length for raw signals
SAMPLE_K_SEGMENTS = 200 # Default number of segments to sample per group for UMAP
DEFAULT_NUM_DATASETS_TO_DISPLAY = 18 # Default number of datasets for plotting if selection is used
# --- 1. Metadata Loading and Dataset Selection ---
def load_and_select_metadata(excel_path, num_datasets_to_display=None, reports_dir="reports"):
"""
Loads metadata, extracts Dataset_Name, sorts by Id, and optionally selects a subset of datasets.
Saves the selected metadata.
"""
try:
metadata_df = pd.read_excel(excel_path)
print(f"Successfully loaded metadata: {excel_path}")
except FileNotFoundError:
print(f"ERROR: Metadata file not found {excel_path}"); return None
except Exception as e:
print(f"ERROR loading metadata: {e}"); return None
if 'Id' not in metadata_df.columns:
print("ERROR: 'Id' column not found in metadata."); return None
if 'Name' not in metadata_df.columns:
print("WARNING: 'Name' column not found. Using 'Dataset_id' as 'Dataset_Name'.")
metadata_df['Dataset_Name'] = metadata_df['Dataset_id'].astype(str)
else:
metadata_df['Dataset_Name'] = metadata_df['Name'].apply(
lambda x: x.split('_')[-1] if isinstance(x, str) and '_' in x else str(x)
)
metadata_df['Id'] = pd.to_numeric(metadata_df['Id'], errors='coerce')
metadata_df.dropna(subset=['Id'], inplace=True)
metadata_df['Id'] = metadata_df['Id'].astype(int)
metadata_df.sort_values(by='Id', inplace=True)
selected_metadata_df = metadata_df.copy()
if num_datasets_to_display is not None and num_datasets_to_display > 0:
# Simple selection: take datasets with the most entries, up to num_datasets_to_display
# More sophisticated selection might be needed (e.g., specific names)
top_datasets = selected_metadata_df['Dataset_Name'].value_counts().nlargest(num_datasets_to_display).index
selected_metadata_df = selected_metadata_df[selected_metadata_df['Dataset_Name'].isin(top_datasets)]
print(f"Selected top {len(top_datasets)} datasets for processing: {top_datasets.tolist()}")
os.makedirs(reports_dir, exist_ok=True)
selected_metadata_path = os.path.join(reports_dir, "selected_metadata.csv")
selected_metadata_df.to_csv(selected_metadata_path, index=False)
print(f"Selected metadata saved to: {selected_metadata_path}")
return selected_metadata_df
# --- 2. Raw Data Loading, Truncation, and Sampling ---
def _process_single_raw_segment(h5_key_meta_tuple, h5_filepath, truncate_m_points):
"""
Loads a single raw data segment from HDF5, truncates/pads it, and returns with metadata.
"""
h5_key, meta_info = h5_key_meta_tuple
try:
with h5py.File(h5_filepath, 'r') as h5f_proc:
raw_data = h5f_proc[h5_key][()]
if not np.issubdtype(raw_data.dtype, np.number): return None
raw_data = raw_data.flatten()
if raw_data.size == 0: return None
if raw_data.size > truncate_m_points:
processed_data = raw_data[:truncate_m_points]
elif raw_data.size < truncate_m_points:
# Pad with zeros if shorter
processed_data = np.pad(raw_data, (0, truncate_m_points - raw_data.size), 'constant')
else:
processed_data = raw_data
return {
'Id': meta_info['Id'],
'H5_Key': h5_key,
'Dataset_id': meta_info.get('Dataset_id', 'N/A'),
'Dataset_Name': meta_info.get('Dataset_Name', 'N/A'),
'Label': meta_info.get('Label', 'N/A'),
'Domain_id': meta_info.get('Domain_id', 'N/A'),
'Fault_level': meta_info.get('Fault_level', 'N/A'),
'raw_segment': processed_data
}
except Exception as e:
print(f" ERROR processing raw segment Id {meta_info.get('Id', 'Unknown')} (H5 key: {h5_key}): {e}")
return None
def load_process_sample_raw_data(h5_filepath, selected_metadata_df, cache_dir,
truncate_m_points=TRUNCATE_M_POINTS,
sample_k_segments=SAMPLE_K_SEGMENTS,
max_workers=None):
"""
Loads raw data based on selected_metadata_df, truncates/pads, samples k segments per group,
and prepares data for UMAP. Caches results.
"""
if selected_metadata_df is None or selected_metadata_df.empty:
print("No metadata provided for raw data processing."); return None, None
os.makedirs(cache_dir, exist_ok=True)
raw_segments_matrix_cache_path = os.path.join(cache_dir, f"sampled_raw_segments_m{truncate_m_points}_k{sample_k_segments}.npy")
raw_segments_labels_cache_path = os.path.join(cache_dir, f"sampled_raw_segments_labels_m{truncate_m_points}_k{sample_k_segments}.csv")
if os.path.exists(raw_segments_matrix_cache_path) and os.path.exists(raw_segments_labels_cache_path):
try:
print("Loading sampled raw data from cache...")
sampled_raw_segments_matrix = np.load(raw_segments_matrix_cache_path)
sampled_labels_df = pd.read_csv(raw_segments_labels_cache_path)
for col in ['Dataset_Name', 'Label', 'Domain_id', 'Fault_level', 'Id', 'H5_Key', 'Dataset_id']:
if col in sampled_labels_df.columns: sampled_labels_df[col] = sampled_labels_df[col].astype(str)
print("Successfully loaded sampled raw data from cache.")
return sampled_raw_segments_matrix, sampled_labels_df
except Exception as e:
print(f"Error loading from cache: {e}. Re-processing.")
if os.path.exists(raw_segments_matrix_cache_path): os.remove(raw_segments_matrix_cache_path)
if os.path.exists(raw_segments_labels_cache_path): os.remove(raw_segments_labels_cache_path)
print("\nProcessing raw data: Loading, Truncating/Padding...")
all_processed_segments_info = []
selected_metadata_df['Id_str'] = selected_metadata_df['Id'].astype(str)
items_to_process_raw = []
h5_keys_not_found_in_h5 = []
try:
with h5py.File(h5_filepath, 'r') as h5f:
h5_keys_in_file = set(h5f.keys())
for _, meta_row in selected_metadata_df.iterrows():
h5_key_candidate = meta_row['Id_str']
if h5_key_candidate in h5_keys_in_file:
items_to_process_raw.append((h5_key_candidate, meta_row.to_dict()))
else:
h5_keys_not_found_in_h5.append(h5_key_candidate)
if h5_keys_not_found_in_h5:
print(f"WARNING: {len(h5_keys_not_found_in_h5)} Ids from metadata not found as keys in HDF5: {h5_keys_not_found_in_h5[:5]}...")
except FileNotFoundError:
print(f"ERROR: HDF5 file {h5_filepath} not found."); return None, None
except Exception as e:
print(f"ERROR opening HDF5 file {h5_filepath}: {e}"); return None, None
if not items_to_process_raw:
print("No items to process from HDF5 based on selected metadata."); return None, None
with concurrent.futures.ProcessPoolExecutor(max_workers=max_workers) as executor:
task_fn = partial(_process_single_raw_segment, h5_filepath=h5_filepath, truncate_m_points=truncate_m_points)
results = list(executor.map(task_fn, items_to_process_raw))
all_processed_segments_info = [res for res in results if res is not None]
if not all_processed_segments_info:
print("No valid raw segments processed."); return None, None
full_segments_df = pd.DataFrame(all_processed_segments_info)
full_segments_df['Id'] = pd.to_numeric(full_segments_df['Id']) # For sorting
full_segments_df.sort_values(by='Id', inplace=True)
print(f"\nSampling {sample_k_segments} segments per (Dataset_Name, Domain_id, Label) group...")
grouping_cols_sample = [col for col in ['Dataset_Name', 'Domain_id', 'Label'] if col in full_segments_df.columns]
if not grouping_cols_sample:
print("WARNING: Not enough columns for grouped sampling. Using all processed segments.")
sampled_df = full_segments_df
else:
def sample_group_k(group):
return group.sample(n=min(len(group), sample_k_segments), random_state=1) # Consistent sampling
sampled_df = full_segments_df.groupby(grouping_cols_sample, group_keys=False).apply(sample_group_k)
sampled_df.sort_values(by='Id', inplace=True) # Ensure final sort by Id
print(f"Sampling complete. Selected {len(sampled_df)} segments for UMAP.")
if sampled_df.empty:
print("No segments remaining after sampling."); return None, None
sampled_raw_segments_matrix = np.array(sampled_df['raw_segment'].tolist())
sampled_labels_df = sampled_df.drop(columns=['raw_segment'])
# Cache the sampled data
try:
np.save(raw_segments_matrix_cache_path, sampled_raw_segments_matrix)
# Ensure string types for labels before saving
for col in ['Dataset_Name', 'Label', 'Domain_id', 'Fault_level', 'Id', 'H5_Key', 'Dataset_id']:
if col in sampled_labels_df.columns: sampled_labels_df[col] = sampled_labels_df[col].astype(str)
sampled_labels_df.to_csv(raw_segments_labels_cache_path, index=False)
print("Sampled raw data cached successfully.")
except Exception as e:
print(f"Error caching sampled raw data: {e}")
return sampled_raw_segments_matrix, sampled_labels_df
# --- 3. UMAP Processing ---
def run_umap_reduction(data_matrix, cache_dir, umap_params, filename_prefix="umap_embedding"):
"""Runs UMAP and caches the embedding."""
n_neighbors = umap_params.get('n_neighbors', 15)
min_dist = umap_params.get('min_dist', 0.1)
n_components = umap_params.get('n_components', 2)
metric = umap_params.get('metric', 'euclidean') # Common for raw signals
random_state = umap_params.get('random_state', 42)
embedding_cache_path = os.path.join(cache_dir, f"{filename_prefix}_nn{n_neighbors}_md{min_dist}_c{n_components}_{metric}.npy")
if os.path.exists(embedding_cache_path):
try:
print("Loading UMAP embedding from cache...")
embedding = np.load(embedding_cache_path)
print("Successfully loaded UMAP embedding from cache.")
return embedding
except Exception as e:
print(f"Error loading UMAP embedding from cache: {e}. Re-calculating.")
if os.path.exists(embedding_cache_path): os.remove(embedding_cache_path)
print(f"\nRunning UMAP (n_neighbors={n_neighbors}, min_dist={min_dist}, metric={metric})...")
if data_matrix.shape[0] <= n_neighbors:
print(f"WARNING: Number of samples ({data_matrix.shape[0]}) is less than or equal to n_neighbors ({n_neighbors}). Adjusting n_neighbors.")
n_neighbors = max(1, data_matrix.shape[0] -1) # UMAP needs n_neighbors < n_samples
if n_neighbors == 0 and data_matrix.shape[0] == 1: # Cannot run UMAP on 1 sample
print("ERROR: Cannot run UMAP on a single sample.")
return None
reducer = umap.UMAP(
n_neighbors=n_neighbors,
min_dist=min_dist,
n_components=n_components,
metric=metric,
random_state=random_state,
verbose=True
)
try:
embedding = reducer.fit_transform(data_matrix)
print("UMAP reduction complete.")
np.save(embedding_cache_path, embedding)
print(f"UMAP embedding cached: {embedding_cache_path}")
return embedding
except Exception as e:
print(f"ERROR during UMAP: {e}"); return None
# --- 4. Plotting ---
def save_plot_multiformat(fig, output_dir, filename_base, high_dpi=300):
"""Saves the given figure in multiple formats."""
formats = {"png": {"dpi": high_dpi}, "svg": {}, "pdf": {}}
for ext, options in formats.items():
try:
save_path = os.path.join(output_dir, f"{filename_base}.{ext}")
fig.savefig(save_path, **options, bbox_inches='tight')
print(f"Plot saved: {save_path}")
except Exception as e:
print(f"ERROR saving plot {save_path}: {e}")
def plot_umap_global(umap_df, hue_col, output_dir, filename_base, title_suffix):
"""Plots the global UMAP, colored by hue_col, and returns the palette."""
print(f"\nPlotting Global UMAP colored by {hue_col}...")
if umap_df is None or umap_df.empty: print("No data for global UMAP plot."); return None
unique_hues = sorted(umap_df[hue_col].unique())
palette = sns.color_palette("husl", n_colors=len(unique_hues)) # husl is good for many distinct colors
# palette = sns.color_palette("viridis", n_colors=len(unique_hues))
color_map = dict(zip(unique_hues, palette))
fig, ax = plt.subplots(figsize=(16, 12))
sns.set_style("whitegrid")
sns.scatterplot(
x='UMAP1', y='UMAP2', hue=hue_col, hue_order=unique_hues,
palette=color_map, data=umap_df, s=50, alpha=0.7, ax=ax,
legend="auto" if len(unique_hues) <= 20 else False
)
ax.set_title(f'Global UMAP: {title_suffix}', fontsize=18)
ax.set_xlabel('UMAP Component 1', fontsize=14)
ax.set_ylabel('UMAP Component 2', fontsize=14)
if len(unique_hues) > 1 and len(unique_hues) <= 20:
ax.legend(title=hue_col, bbox_to_anchor=(1.02, 1), loc='upper left', borderaxespad=0.)
save_plot_multiformat(fig, output_dir, filename_base)
plt.show()
plt.close(fig)
return color_map
def plot_umap_faceted_by_dataset(umap_df, facet_col, color_col, style_col, dataset_color_map, output_dir, filename_base_prefix, title_prefix):
"""
Plots UMAP data faceted by facet_col (e.g., Dataset_Name).
Markers are colored by color_col (e.g., Domain_id) and styled by style_col (e.g., Label).
Subplot frames are colored based on dataset_color_map.
"""
print(f"\nPlotting Faceted UMAP: Facet by {facet_col}, Color by {color_col}, Style by {style_col}...")
if umap_df is None or umap_df.empty: print(f"No data for faceted UMAP plot ({title_prefix})."); return
if not all(c in umap_df.columns for c in [facet_col, color_col, style_col]):
print(f"ERROR: Required columns '{facet_col}', '{color_col}', or '{style_col}' not in DataFrame for faceted plot."); return
unique_facets = sorted(umap_df[facet_col].unique())
if not unique_facets: print("No unique facets found."); return
# Define consistent color and style mappings across all facets
all_unique_colors = sorted(umap_df[color_col].astype(str).unique())
color_palette_for_plot = sns.color_palette("Set2", n_colors=len(all_unique_colors))
all_unique_styles = sorted(umap_df[style_col].astype(str).unique())
# Define a list of markers - MODIFIED TO USE ONLY FILLED MARKERS or compatible ones
available_markers = ['o', 's', 'D', '^', 'v', '<', '>', 'p', '*', 'h', 'H'] # Removed '+', 'X', 'P', '.' which can cause issues
# 'X' and 'P' are often filled but can be treated differently
# 'p' is pentagon, '*' is star, 'h'/'H' are hexagons
if len(all_unique_styles) > len(available_markers):
print(f"WARNING: More unique styles ({len(all_unique_styles)}) than available distinct markers ({len(available_markers)}). Markers will repeat.")
markers_for_plot = {style_val: available_markers[j % len(available_markers)] for j, style_val in enumerate(all_unique_styles)}
n_facets = len(unique_facets)
n_cols = min(3, n_facets)
n_rows = (n_facets + n_cols - 1) // n_cols
fig, axes = plt.subplots(n_rows, n_cols, figsize=(6 * n_cols, 5 * n_rows), squeeze=False, sharex=True, sharey=True)
axes_flat = axes.flatten()
sns.set_style("whitegrid")
for i, facet_value in enumerate(unique_facets):
if i >= len(axes_flat): break
ax = axes_flat[i]
facet_data = umap_df[umap_df[facet_col] == facet_value].copy()
facet_data[color_col] = facet_data[color_col].astype(str)
facet_data[style_col] = facet_data[style_col].astype(str)
if facet_data.empty:
ax.text(0.5, 0.5, "No data", ha='center', va='center', transform=ax.transAxes)
ax.set_title(f"{facet_value}", fontsize=12, color='grey')
ax.tick_params(labelbottom=False, labelleft=False)
continue
# Ensure style_order matches the keys in markers_for_plot for consistency
current_style_order = [s for s in all_unique_styles if s in facet_data[style_col].unique()]
sns.scatterplot(
x='UMAP1', y='UMAP2',
hue=color_col,
hue_order=all_unique_colors,
palette=color_palette_for_plot,
style=style_col,
style_order=current_style_order, # Use the order of styles present in the current facet_data
markers=markers_for_plot,
data=facet_data,
s=70, alpha=0.85, ax=ax, # Increased marker size slightly
legend='auto'
)
ax.set_title(f"{facet_value}", fontsize=14, fontweight='bold', color=dataset_color_map.get(facet_value, 'black'))
if n_rows > 1 and i < (n_rows - 1) * n_cols :
ax.set_xlabel('')
ax.tick_params(labelbottom=False)
if n_cols > 1 and i % n_cols != 0:
ax.set_ylabel('')
ax.tick_params(labelleft=False)
spine_color = dataset_color_map.get(facet_value, 'grey')
for spine in ax.spines.values():
spine.set_edgecolor(spine_color)
spine.set_linewidth(2.5)
try:
handles, labels = ax.get_legend_handles_labels()
if handles:
legend_title = f"{color_col} (Color)\n{style_col} (Marker)"
# Filter legend items to only those present in the current facet_data
# This is a bit more complex as seaborn might already do this.
# The primary goal is to ensure the legend isn't overly crowded.
num_legend_items = len(handles)
if num_legend_items > 10:
ax.legend(title=legend_title, fontsize='xx-small', loc='best', ncol=2 if num_legend_items > 5 else 1)
if num_legend_items > 20:
ax.legend().set_visible(False)
elif num_legend_items > 0 : # Only show legend if there are items
ax.legend(title=legend_title, fontsize='x-small', loc='best')
else: # No items, hide legend explicitly
if ax.get_legend() is not None:
ax.get_legend().set_visible(False)
except AttributeError:
pass
for j in range(i + 1, len(axes_flat)):
fig.delaxes(axes_flat[j])
fig.suptitle(f"{title_prefix}\n(Faceted by {facet_col})", fontsize=20, y=0.99)
if n_rows > 0 and n_cols > 0 :
fig.supxlabel("UMAP Component 1", fontsize=16, y=0.01 if n_rows >1 else -0.02)
fig.supylabel("UMAP Component 2", fontsize=16, x=0.01 if n_cols >1 else -0.02)
plt.tight_layout(rect=[0.03, 0.03, 0.97, 0.96])
save_plot_multiformat(fig, output_dir, f"{filename_base_prefix}_facet_{facet_col}_color_{color_col}_style_{style_col}")
plt.show()
plt.close(fig)
# --- 5. Main Orchestration ---
def main():
# --- File Paths & Directories ---
excel_filepath = "/home/user/data/PHMbenchdata/PHM-Vibench/metadata_6_1.xlsx" # Ensure this is correct
h5_filepath = "/home/user/data/PHMbenchdata/PHM-Vibench/metadata_6_1.h5" # Ensure this is correct
base_output_dir = f"analysis_raw_m{TRUNCATE_M_POINTS}_k{SAMPLE_K_SEGMENTS}"
cache_dir = os.path.join(base_output_dir, "00_cache")
reports_dir = os.path.join(base_output_dir, "01_reports")
umap_plots_dir = os.path.join(base_output_dir, "02_umap_plots")
for d in [cache_dir, reports_dir, umap_plots_dir]:
os.makedirs(d, exist_ok=True)
# --- Parameters ---
num_datasets_to_plot = DEFAULT_NUM_DATASETS_TO_DISPLAY # Or None for all, or specific list
max_parallel_workers = os.cpu_count() // 2 if os.cpu_count() and os.cpu_count() > 1 else 1
umap_parameters = {
'n_neighbors': 200, # Might need tuning for raw signals
'min_dist': 0.1,
'n_components': 2,
'metric': 'correlation', # 'correlation' or 'dtw' could be alternatives for time series
'random_state': 42
}
# --- Step 1: Load and Select Metadata ---
# --- Step 1: Load and Select Metadata ---
print("--- Step 1: Loading and Selecting Metadata ---")
selected_metadata = load_and_select_metadata(excel_filepath,
num_datasets_to_display=num_datasets_to_plot,
reports_dir=reports_dir)
if selected_metadata is None: print("Analysis halted due to metadata loading issues."); return
# --- Step 2: Load, Process, and Sample Raw Data ---
print("\n--- Step 2: Loading, Processing, and Sampling Raw Data ---")
sampled_raw_matrix, sampled_labels = load_process_sample_raw_data(
h5_filepath, selected_metadata, cache_dir,
truncate_m_points=TRUNCATE_M_POINTS,
sample_k_segments=SAMPLE_K_SEGMENTS,
max_workers=max_parallel_workers
)
if sampled_raw_matrix is None or sampled_labels is None or sampled_raw_matrix.shape[0] == 0:
print("Analysis halted: No usable raw data segments after processing/sampling."); return
print(f"Shape of sampled raw data matrix for UMAP: {sampled_raw_matrix.shape}")
print(f"Number of labels for UMAP: {len(sampled_labels)}")
# --- Step 3: Run UMAP ---
print("\n--- Step 3: Running UMAP ---")
umap_embedding = run_umap_reduction(sampled_raw_matrix, cache_dir, umap_parameters,
filename_prefix=f"raw_m{TRUNCATE_M_POINTS}_k{SAMPLE_K_SEGMENTS}")
if umap_embedding is None:
print("Analysis halted: UMAP embedding failed."); return
# Create DataFrame for plotting
umap_plot_df = sampled_labels.copy()
umap_plot_df['UMAP1'] = umap_embedding[:, 0]
umap_plot_df['UMAP2'] = umap_embedding[:, 1]
# Save UMAP results with labels
umap_results_path = os.path.join(reports_dir, f"umap_results_m{TRUNCATE_M_POINTS}_k{SAMPLE_K_SEGMENTS}.csv")
umap_plot_df.to_csv(umap_results_path, index=False)
print(f"UMAP results with labels saved to: {umap_results_path}")
# --- Step 4: Plotting ---
print("\n--- Step 4: Generating UMAP Plots ---")
# Global plot by Dataset_Name
dataset_color_palette = plot_umap_global(
umap_plot_df,
hue_col='Dataset_Name',
output_dir=umap_plots_dir,
filename_base="global_umap_by_dataset_name",
title_suffix="Colored by Dataset Name"
)
if dataset_color_palette is None: dataset_color_palette = {} # Fallback
# Combined Faceted plot:
# Facet by Dataset_Name, Color by Domain_id, Marker Style by Label
plot_umap_faceted_by_dataset(
umap_plot_df,
facet_col='Dataset_Name',
color_col='Domain_id', # Color by Domain_id
style_col='Label', # Marker style by Label
dataset_color_map=dataset_color_palette, # For subplot frame color
output_dir=umap_plots_dir,
filename_base_prefix="faceted_umap_combined", # Updated filename prefix
title_prefix="UMAP: Domain ID (Color) & Label (Marker)" # Updated title
)
print("\n--- Analysis Complete ---")
if __name__ == '__main__':
main()