v1.4.0

changelog/v1_4_0.md

@@ -0,0 +1,12 @@
+## v1.4.0
+### Features
+#### Frequency reduction
+A custom function can now be provided to reduce the array of frequency of each
+track into a single frequency. This is mainly intended to provide a way to
+switch between mean and median
+#### Settable properties
+Learned properties (frequencies, decays, amplitudes) can now be setted. This
+allows post-fit manipulations, e.g. pitch-shift
+### Improvements
+Now forcing correct package version in notebooks. Also using quiet mode instead
+of clearing output

docs/source/changelog.rst

@@ -1,6 +1,7 @@
 Changelog
 =========
 
+.. mdinclude:: ../../changelog/v1_4_0.md
 .. mdinclude:: ../../changelog/v1_3_0.md
 .. mdinclude:: ../../changelog/v1_2_3.md
 .. mdinclude:: ../../changelog/v1_2_2.md

notebooks/SAMPLE.ipynb

@@ -29,11 +29,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Install requirements\n",
-    "from IPython.display import clear_output\n",
     "import sys\n",
-    "!$sys.executable -m pip install lim-sample[notebooks,plots]==1.3.0\n",
-    "clear_output()"
+    "!$sys.executable -m pip install -qU lim-sample[notebooks,plots]==1.4.0"
    ]
   },
   {
@@ -331,6 +328,42 @@
     "resize(12, 12)\n",
     "play(x_dual, rate=fs)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Manipulate\n",
+    "We can also manipulate the estimated modal parameters to generate a different sound!\n",
+    "\n",
+    "As an example we'll tune up the pitches, power-transform the decays and level the amplitudes"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import copy\n",
+    "sample_mod = copy.deepcopy(sample)\n",
+    "\n",
+    "semitones = 1  #@param {type:\"slider\", min:-12, max:12, step:1}\n",
+    "gamma = 3.14  #@param {type:\"slider\", min:0.01, max:4, step:0.01}\n",
+    "\n",
+    "sample_mod.freqs_ *= 2**(semitones/12)\n",
+    "sample_mod.decays_ **= gamma\n",
+    "sample_mod.amps_ = np.full(\n",
+    "  sample.amps_.size,\n",
+    "  1/sample.amps_.size\n",
+    ")\n",
+    "\n",
+    "x_mod = sample_mod.predict(np.arange(x.size) / fs)\n",
+    "waveplot(x_mod, sr=fs, alpha=.5, zorder=100)\n",
+    "plt.grid()\n",
+    "resize()\n",
+    "play(x_mod, rate=fs)"
+   ]
   }
  ],
  "metadata": {

notebooks/SDT.ipynb

@@ -30,10 +30,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from IPython.display import clear_output\n",
     "import sys\n",
-    "!$sys.executable -m pip install lim-sample[notebooks,plots]==1.3.0\n",
-    "clear_output()"
+    "!$sys.executable -m pip install -qU lim-sample[notebooks,plots]==1.4.0"
    ]
   },
   {

sample/__init__.py

@@ -25,7 +25,7 @@
 from sample.sample import SAMPLE
 
 
-__version__ = "1.3.0"
+__version__ = "1.4.0"
 
 
 @cli.main(__name__)

sample/sample.py

@@ -5,6 +5,7 @@
 import numpy as np
 import functools
 import copy
+from typing import Callable
 
 
 class SAMPLE(base.RegressorMixin, base.BaseEstimator):
@@ -19,6 +20,8 @@ class SAMPLE(base.RegressorMixin, base.BaseEstimator):
       coefficient of :data:`regressor`
     regressor_q (str): Attribute name for the estimated intercept
       coefficient of :data:`regressor`
+    freq_reduce (callable): Callable function for reducing the frequency track
+      into a single frequency. Defaults to :func:`numpy.mean`
     **kwargs: Keyword arguments, will be set as parameters of submodels. For a
       complete list of all parameter names and default values, please, run
       :data:`SAMPLE().get_params()`. For an explanation of the parameters,
@@ -29,12 +32,14 @@ def __init__(
     regressor=HingeRegression(),
     regressor_k: str = "k_",
     regressor_q: str = "q_",
+    freq_reduce: Callable[[np.ndarray], float] = np.mean,
     **kwargs,
   ):
     self.sinusoidal_model = sinusoidal_model
     self.regressor = regressor
     self.regressor_k = regressor_k
     self.regressor_q = regressor_q
+    self.freq_reduce = freq_reduce
     self.set_params(**kwargs)
 
   @property
@@ -68,7 +73,7 @@ def fit(self, x: np.ndarray, y=None, **kwargs):
     self.param_matrix_ = np.zeros((3, len(tracks)))
     for i, t in enumerate(tracks):
       notnans = np.logical_not(np.isnan(t["mag"]))
-      self.param_matrix_[0, i] = np.mean(t["freq"][notnans])
+      self.param_matrix_[0, i] = self.freq_reduce(t["freq"][notnans])
       x_ = (t["start_frame"] + np.arange(t["mag"].size)[notnans]) * \
            self.sinusoidal_model.h / self.sinusoidal_model.fs
       y_ = t["mag"][notnans]
@@ -86,16 +91,28 @@ def freqs_(self) -> np.ndarray:
     """Learned modal frequencies"""
     return self.param_matrix_[0, :]
 
+  @freqs_.setter
+  def freqs_(self, f: np.ndarray):
+    self.param_matrix_[0, :] = f
+
   @property
   def decays_(self) -> np.ndarray:
     """Learned modal decays"""
     return self.param_matrix_[1, :]
 
+  @decays_.setter
+  def decays_(self, d: np.ndarray):
+    self.param_matrix_[1, :] = d
+
   @property
   def amps_(self) -> np.ndarray:
     """Learned modal amplitudes"""
     return self.param_matrix_[2, :]
 
+  @amps_.setter
+  def amps_(self, a: np.ndarray):
+    self.param_matrix_[2, :] = a
+
   @property
   def energies_(self) -> np.ndarray:
     """Learned modal energies"""

tests/test_sample.py

@@ -138,6 +138,39 @@ def test_strip_reverse(self):
     with self.assert_doesnt_raise():
       s.fit(self.x)
 
+  def test_freqs_modification(self):
+    """Test that frequencies can be effectively manipulated post-fit"""
+    s = copy.deepcopy(self.sample)
+    s.fit(self.x)
+    t = copy.deepcopy(s)
+    k = 1.2
+    t.freqs_ *= k
+    for i, (f_s, f_t) in enumerate(zip(s.freqs_, t.freqs_)):
+      with self.subTest(partial=i):
+        self.assertEqual(f_s * k, f_t)
+
+  def test_decays_modification(self):
+    """Test that decays can be effectively manipulated post-fit"""
+    s = copy.deepcopy(self.sample)
+    s.fit(self.x)
+    t = copy.deepcopy(s)
+    k = 1.2
+    t.decays_ *= k
+    for i, (d_s, d_t) in enumerate(zip(s.decays_, t.decays_)):
+      with self.subTest(partial=i):
+        self.assertEqual(d_s * k, d_t)
+
+  def test_amps_modification(self):
+    """Test that amplitudes can be effectively manipulated post-fit"""
+    s = copy.deepcopy(self.sample)
+    s.fit(self.x)
+    t = copy.deepcopy(s)
+    k = 1.2
+    t.amps_ *= k
+    for i, (a_s, a_t) in enumerate(zip(s.amps_, t.amps_)):
+      with self.subTest(partial=i):
+        self.assertEqual(a_s * k, a_t)
+
   def test_plot_2d(self):
     """Test 2D plot"""
     s = copy.deepcopy(self.sample).fit(self.x)