Again, numerous changes.

Mostly figure polishing and fixing.
Crucial fix to "short" invariance analysis.

python/save_field_data.py

@@ -0,0 +1,62 @@
+import numpy as np
+from thunderhopper.filetools import search_files, crop_paths
+from thunderhopper.model import configuration, process_signal
+from thunderhopper.modeltools import load_data
+from IPython import embed
+
+## SETTINGS:
+
+# General:
+search_target = '*'
+mode = ['song', 'noise'][1]
+input_folder = f'../data/field/raw/{mode}/'
+output_folder = f'../data/field/processed/{mode}/'
+stages = ['raw', 'norm']
+if False:
+    # Overwrites edited:
+    stages.append('songs')
+
+# Interactivity:
+reload_saved = False
+gui = True
+
+# Processing:
+env_rate = 96000.0
+feat_rate = 96000.0
+sigmas = [0.001, 0.002, 0.004, 0.008, 0.016, 0.032]
+types = [1, -1, 2, -2, 3, -3, 4, -4, 5, -5,
+         6, -6, 7, -7, 8, -8, 9, -9, 10, -10]
+config = configuration(env_rate, feat_rate, types=types, sigmas=sigmas)
+config.update({
+    'channel': None,
+    'rate_ratio': None,
+    'env_fcut': 250,
+    'db_ref': 1,
+    'inv_fcut': 10,
+    'feat_thresh': np.load('../data/kernel_thresholds.npy') * 0.2,
+    'feat_fcut': 0.5,
+    'label_channels': np.array([0]),
+    'label_thresh': 0.5,
+    })
+
+## PREPARATION:
+
+# Fetch WAV recording files:
+input_paths = search_files(search_target, ext='wav', dir=input_folder)
+path_names = crop_paths(input_paths)
+
+# PROCESSING:
+
+# Run processing pipeline:
+for path, name in zip(input_paths, path_names):
+    print('Processing:', name)
+
+    # Fetch and store representations:
+    save = None if output_folder is None else output_folder + f'{name}.npz'
+    process_signal(config, stages, path, save=save, label_edit=gui)
+
+    # Cross-control:
+    if reload_saved:
+        data, params = load_data(save, stages, ['songs'])
+        embed()
+print('Done.')