Holiday syncing :)

python/save_ref_measures.py

@@ -0,0 +1,65 @@
+import numpy as np
+from thunderhopper.filters import decibel, sosfilter
+from thunderhopper.model import convolve_kernels, process_signal
+from thunderhopper.modeltools import load_data
+from IPython import embed
+
+## SETTINGS:
+
+# General:
+mode = ['log_hp', 'thresh_lp', 'full'][2]
+noise_path = '../data/processed/white_noise_sd-1.npz'
+save_path = '../data/inv/'
+pad = np.array([0.1, 0.9])
+
+stages = dict(
+    log_hp=['filt', 'env', 'log', 'inv'],
+    thresh_lp=['inv', 'conv', 'feat'],
+    full=['raw', 'filt', 'env', 'log', 'inv', 'conv', 'feat']
+)[mode]
+
+# PROCESSING:
+
+print(f'Fetching references for {mode} invariance...')
+
+# Load pure-noise starter representation:
+noise_data, config = load_data(noise_path, stages[0])
+starter = noise_data[stages[0]]
+
+# Prepare buffered measurement segment:
+pad = (pad * starter.shape[0]).astype(int)
+segment = np.arange(starter.shape[0])[pad[0]:pad[1]]
+
+# Normalize starter:
+starter /= starter[segment].std()
+
+# Run pipeline:
+if mode == 'log_hp':
+    data = {'filt': starter}
+    data['env'] = sosfilter(np.abs(data['filt']), config['rate'], config['env_fcut'],
+                            'lp', padtype='even', padlen=config['padlen'])
+    data['log'] = decibel(data['env'], ref=1)
+    data['inv'] = sosfilter(data['log'], config['env_rate'], config['inv_fcut'],
+                            'hp', padtype='constant', padlen=config['padlen'])
+elif mode == 'thresh_lp':
+    data = {'inv': starter}
+    data['conv'] = convolve_kernels(data['inv'], config['kernels'], config['k_specs'])
+    data['feat'] = sosfilter((data['conv'] > config['feat_thresh']).astype(float),
+                             config['env_rate'], config['feat_fcut'], 'lp',
+                             padtype='fixed', padlen=config['padlen'])
+elif mode == 'full':
+    data = process_signal(config, stages, signal=starter, rate=config['rate'])[0]
+
+# Get measures:
+measures = {}
+for stage in stages:
+    if stage == 'feat':
+        measures[stage] = data[stage][segment, :].mean(axis=0)
+    else:
+        measures[stage] = data[stage][segment, ...].std(axis=0)
+
+# Save results:
+np.savez(save_path + f'{mode}/ref_measures.npz', **measures)
+
+print('Done.')
+embed()