add scale option to bank compression along with higher order interpolation functions (2nd, 3rd, and 4th order). (#5218)

* add scale option to bank compression to make initial points more course

* fixes

* stay with this for now

* another fix

* add inline quadratic interpolation

* add cubic inline

* also add quartic

* remove boilerplate

* set scale to 1 as default

* try this for mac compmlication

* minor performance improvement

Author: ahnitz

bin/pycbc_compress_bank

@@ -22,6 +22,7 @@ hdf file."""
 import argparse
 import numpy
 import logging
+import time
 
 import pycbc
 from pycbc import psd
@@ -87,6 +88,11 @@ parser.add_argument("--precision", type=str, choices=["double", "single"],
 parser.add_argument("--force", action="store_true", default=False,
                     help="Overwrite the given hdf file if it exists. "
                     "Otherwise, an error is raised.")
+parser.add_argument("--scale", type=float, default=1,
+                    help="Scale factor to apply to the courseness of the initial "
+                         "compression points. >1 means a courser set of initial "
+                         "points. The algorithm will still add points to meet"
+                         "the tolerance goals.")
 parser.add_argument(
     "--do-not-compress",
     nargs="+",
@@ -161,7 +167,10 @@ for ii in range(templates.size):
     if tmplt['approximant'] in args.do_not_compress:
         continue
     # generate the waveform
+    t1 = time.time()
     htilde = bank[ii]
+    htime = time.time() - t1
+    logging.info("... %.2f ms to generate normally", htime * 1000)
     fmin=tmplt.f_lower
     template_duration = htilde.chirp_length
     output['template_duration'][ii] = template_duration
@@ -180,11 +189,13 @@ for ii in range(templates.size):
     # get the compressed sample points
     if args.compression_algorithm == 'mchirp':
         sample_points = compress.mchirp_compression(tmplt.mass1, tmplt.mass2,
-            fmin, kmax*df, min_seglen=args.t_pad, df_multiple=df).astype(
+            fmin, kmax*df, min_seglen=args.t_pad, df_multiple=df,
+            scale=args.scale).astype(
             real_same_precision_as(htilde))
     elif args.compression_algorithm == 'spa':
         sample_points = compress.spa_compression(htilde, fmin, kmax*df,
-            min_seglen=args.t_pad).astype(real_same_precision_as(htilde))
+            min_seglen=args.t_pad,
+            scale=args.scale).astype(real_same_precision_as(htilde))
     else:
         raise ValueError("unrecognized compression algorithm %s" %(
             args.compression_algorithm))