Merge pull request #109 from bmcfee/032-tweaks

faster import, no int outputs

.github/workflows/ci-minimal.yml

@@ -34,7 +34,7 @@ jobs:
       shell: bash -l {0}
       run: |
         mamba install typing_extensions!=4.2 pytest
-        mamba install pip numpy==1.17 scipy==1.0 numba==0.47
+        mamba install pip numpy==1.17 scipy==1.0 numba==0.53
 
     - name: Install
       shell: bash -l {0}

docs/changes.rst

@@ -1,6 +1,13 @@
 Changes
 -------
 
+v0.4.0
+~~~~~~
+2022-08-05
+
+- `#109 <https://github.com/bmcfee/resampy/pull/109>`_ Reduced import time and switched to parallel=False by default.
+- `#109 <https://github.com/bmcfee/resampy/pull/109>`_ Integer-valued inputs now produce floating point outputs.
+
 v0.3.1
 ~~~~~~
 2022-07-07

docs/example.rst

@@ -72,6 +72,31 @@ The next block illustrates resampling along an arbitrary dimension.
 
     # y_low is now a 10-by-3-(5*11025)-by-2 tensor of data
 
+Integer-valued samples
+======================
+Integer-valued inputs are supported, but because resampy interpolates between
+sample values, it will always produce a floating-point output.
+If you really need integer-valued outputs after resampling, you'll have to cast the
+output array as demonstrated below.
+
+.. code-block:: python
+   :linenos:
+
+    import numpy as np
+    import resampy
+
+    sr_orig = 22050
+
+    # Create 5 seconds of random integer noise
+    x = np.random.randint(-32768, high=32767, size=5*sr_orig, dtype=np.int16)
+
+    # resample, y will be floating-point type
+    y = resampy.resample(x, sr_orig, 11025)
+
+    # Cast back to match x's dtype
+    y_int = y.astype(x.dtype)
+
+
 Advanced filtering
 ==================
 resampy allows you to control the design of the filters used in resampling operations.
@@ -100,28 +125,3 @@ resampy allows you to control the design of the filters used in resampling opera
     y = resampy.resample(x, sr_orig, sr_new, filter='kaiser_fast')
 
 
-Benchmarking
-============
-Benchmarking `resampy` is relatively simple, using `ipython`'s ``%timeit`` magic.
-The following example demonstrates resampling a monophonic signal of 400000 samples from
-22.05 KHz to 16 KHz using both `resampy` and `scipy.signal.resample`.
-
-.. code-block:: python
-
-    In [1]: import numpy as np
-
-    In [2]: import scipy
-    
-    In [3]: import resampy
-    
-    In [4]: x = np.random.randn(400000)
-    
-    In [5]: sr_in, sr_out = 22050, 16000
-    
-    In [6]: %timeit resampy.resample(x, sr_in, sr_out, axis=-1)
-    1 loop, best of 3: 199 ms per loop
-    
-    In [7]: %timeit scipy.signal.resample(x,
-       ...:                               int(x.shape[-1] * sr_out / float(sr_in)),
-       ...:                               axis=-1)
-    1 loop, best of 3: 6min 5s per loop

resampy/core.py

@@ -12,7 +12,7 @@
 
 
 def resample(
-    x, sr_orig, sr_new, axis=-1, filter="kaiser_best", parallel=True, **kwargs
+    x, sr_orig, sr_new, axis=-1, filter="kaiser_best", parallel=False, **kwargs
 ):
     """Resample a signal x from sr_orig to sr_new along a given axis.
 
@@ -116,7 +116,12 @@ def resample(
         )
 
     # Preserve contiguity of input (if it exists)
-    y = np.zeros_like(x, shape=shape)
+    if np.issubdtype(x.dtype, np.integer):
+        dtype = np.float32
+    else:
+        dtype = x.dtype
+
+    y = np.zeros_like(x, dtype=dtype, shape=shape)
 
     interp_win, precision, _ = get_filter(filter, **kwargs)
 
@@ -155,7 +160,7 @@ def resample(
 
 
 def resample_nu(
-    x, sr_orig, t_out, axis=-1, filter="kaiser_best", parallel=True, **kwargs
+    x, sr_orig, t_out, axis=-1, filter="kaiser_best", parallel=False, **kwargs
 ):
     """Interpolate a signal x at specified positions (t_out) along a given axis.
 
@@ -239,7 +244,11 @@ def resample_nu(
     shape = list(x.shape)
     shape[axis] = len(t_out)
 
-    y = np.zeros_like(x, shape=shape)
+    if np.issubdtype(x.dtype, np.integer):
+        dtype = np.float32
+    else:
+        dtype = x.dtype
+    y = np.zeros_like(x, dtype=dtype, shape=shape)
 
     interp_win, precision, _ = get_filter(filter, **kwargs)
 

resampy/interpn.py

@@ -1,18 +1,7 @@
 #!/usr/bin/env python
 """Numba implementation of resampler"""
 
-from numba import (
-    guvectorize,
-    float32,
-    float64,
-    jit,
-    prange,
-    int16,
-    int32,
-    int64,
-    complex64,
-    complex128,
-)
+from numba import guvectorize, jit, prange
 
 
 def _resample_loop(x, t_out, interp_win, interp_delta, num_table, scale, y):
@@ -82,31 +71,6 @@ def _resample_loop(x, t_out, interp_win, interp_delta, num_table, scale, y):
 
 
 @guvectorize(
-    [
-        (int16[:], float64[:], float64[:], float64[:], int32, float32, int16[:]),
-        (int32[:], float64[:], float64[:], float64[:], int32, float32, int32[:]),
-        (int64[:], float64[:], float64[:], float64[:], int32, float32, int64[:]),
-        (float32[:], float64[:], float64[:], float64[:], int32, float32, float32[:]),
-        (float64[:], float64[:], float64[:], float64[:], int32, float32, float64[:]),
-        (
-            complex64[:],
-            float64[:],
-            float64[:],
-            float64[:],
-            int32,
-            float32,
-            complex64[:],
-        ),
-        (
-            complex128[:],
-            float64[:],
-            float64[:],
-            float64[:],
-            int32,
-            float32,
-            complex128[:],
-        ),
-    ],
     "(n),(m),(p),(p),(),()->(m)",
     nopython=True,
 )
@@ -115,31 +79,6 @@ def resample_f_p(x, t_out, interp_win, interp_delta, num_table, scale, y):
 
 
 @guvectorize(
-    [
-        (int16[:], float64[:], float64[:], float64[:], int32, float32, int16[:]),
-        (int32[:], float64[:], float64[:], float64[:], int32, float32, int32[:]),
-        (int64[:], float64[:], float64[:], float64[:], int32, float32, int64[:]),
-        (float32[:], float64[:], float64[:], float64[:], int32, float32, float32[:]),
-        (float64[:], float64[:], float64[:], float64[:], int32, float32, float64[:]),
-        (
-            complex64[:],
-            float64[:],
-            float64[:],
-            float64[:],
-            int32,
-            float32,
-            complex64[:],
-        ),
-        (
-            complex128[:],
-            float64[:],
-            float64[:],
-            float64[:],
-            int32,
-            float32,
-            complex128[:],
-        ),
-    ],
     "(n),(m),(p),(p),(),()->(m)",
     nopython=True,
 )

resampy/version.py

@@ -2,5 +2,5 @@
 # -*- coding: utf-8 -*-
 """Version info"""
 
-short_version = '0.3'
-version = '0.3.1'
+short_version = '0.4'
+version = '0.4.0'

setup.cfg

@@ -39,7 +39,7 @@ install_package_data = True
 python_requires >= 3.6
 install_requires =
     numpy>=1.17
-    numba>=0.47
+    numba>=0.53
 
 [options.package_data]
 resampy = data/*

tests/test_core.py

@@ -77,7 +77,7 @@ def test_bad_num_zeros():
 
 
 @pytest.mark.parametrize('dtype', [np.float32, np.float64,
-                                   np.int16, np.int32, np.int64])
+                                   np.complex64, np.complex128])
 def test_dtype(dtype):
     x = np.random.randn(100).astype(dtype)
 
@@ -86,8 +86,26 @@ def test_dtype(dtype):
     assert x.dtype == y.dtype
 
 
+@pytest.mark.parametrize('dtype', [np.int16, np.int32, np.int64])
+def test_dtype_int(dtype):
+    x = (32767 * np.random.randn(100)).astype(dtype)
+
+    y = resampy.resample(x, 100, 200)
+
+    assert y.dtype == np.float32
+
+
+@pytest.mark.parametrize('dtype', [np.int16, np.int32, np.int64])
+def test_dtype_int_nu(dtype):
+    x = (32767 * np.random.randn(100)).astype(dtype)
+    t = np.arange(2 * len(x) - 1) / 2
+
+    y = resampy.resample_nu(x, 1., t)
+    assert y.dtype == np.float32
+
+
 @pytest.mark.parametrize('dtype', [np.float32, np.float64,
-                                   np.int16, np.int32, np.int64])
+                                   np.complex64, np.complex128])
 def test_resample_nu_dtype(dtype):
     x = np.random.randn(100).astype(dtype)
     t = np.arange(2 * len(x) - 1) / 2