Fix NaN handling in Record.adc, and other fixes (#481)

Fix several bugs in `Record.adc`:

1. Previously, the function would try to convert all samples to integers
and then, for any samples that were NaN, replace the corresponding
elements with the appropriate sentinel value. Even though this was
probably safe in most cases, casting NaN to an integer is
implementation-defined behavior, and raises a warning by default (issue
#480).

2. NaN just plain wasn't handled for the `inplace=True, expanded=False`
case. (Currently, we don't use `inplace=True` anywhere internally;
although it saves a bit of memory, it's destructive and so it's probably
wise for high-level functions like `wrsamp` to avoid it.)

3. The `expanded=True` case relied on `self.n_sig` (in contrast to
`expanded=False`, which operates based on the dimensions of `p_signal`.)
This meant it would fail if the caller didn't explicitly set `n_sig`,
which was an annoying inconsistency.

Also, tidy up duplicated code and make things a little more efficient.

A side note: I don't think the `inplace=True` mode is particularly great
to have. It conflates two things (modifying the Record object
attributes, which many applications want; and modifying the array
contents, which you may think you want until you realize it subtly
breaks something.) It does save some memory, but not as much as you'd
hope. (That `copy=False` is pretty much a lie.) And of course I don't
like functions whose return type is dependent on their arguments. So I
would definitely put `inplace` on the chopping block for 5.0.0. Still, I
think the updated code here isn't too terribly ugly.

This set of changes is the first step to making `wfdb.wrsamp` work for
multi-frequency (issue #336). Next is to fix `Record.calc_adc_params`,
then `Record.set_d_features`.

tests/test_record.py

@@ -1053,19 +1053,20 @@ def test_physical_conversion(self):
         adc_gain = [1.0, 1234.567, 765.4321]
         baseline = [10, 20, -30]
         d_signal = np.repeat(np.arange(-100, 100), 3).reshape(-1, 3)
+        d_signal[5:10, :] = [-32768, -2048, -128]
         e_d_signal = list(d_signal.transpose())
-        fmt = ["16", "16", "16"]
+        fmt = ["16", "212", "80"]
 
         # Test adding or subtracting a small offset (0.01 ADU) to check
         # that we correctly round to the nearest integer
         for offset in (0, -0.01, 0.01):
             p_signal = (d_signal + offset - baseline) / adc_gain
+            p_signal[5:10, :] = np.nan
             e_p_signal = list(p_signal.transpose())
 
             # Test converting p_signal to d_signal
 
             record = wfdb.Record(
-                n_sig=n_sig,
                 p_signal=p_signal.copy(),
                 adc_gain=adc_gain,
                 baseline=baseline,
@@ -1081,7 +1082,6 @@ def test_physical_conversion(self):
             # Test converting e_p_signal to e_d_signal
 
             record = wfdb.Record(
-                n_sig=n_sig,
                 e_p_signal=[s.copy() for s in e_p_signal],
                 adc_gain=adc_gain,
                 baseline=baseline,
@@ -1108,7 +1108,7 @@ def test_physical_conversion(self):
                 p_signal=p_signal,
                 adc_gain=adc_gain,
                 baseline=baseline,
-                fmt=["16", "16", "16"],
+                fmt=fmt,
                 write_dir=self.temp_path,
             )
             record = wfdb.rdrecord(

wfdb/io/_signal.py

@@ -532,68 +532,60 @@ def adc(self, expanded=False, inplace=False):
         # To do: choose the minimum return res needed
         intdtype = "int64"
 
+        # Convert a physical (1D or 2D) signal array to digital.  Note that
+        # the input array is modified!
+        def adc_inplace(p_signal, adc_gain, baseline, d_nan):
+            nanlocs = np.isnan(p_signal)
+            np.multiply(p_signal, adc_gain, p_signal)
+            np.add(p_signal, baseline, p_signal)
+            np.round(p_signal, 0, p_signal)
+            np.copyto(p_signal, d_nan, where=nanlocs)
+            d_signal = p_signal.astype(intdtype, copy=False)
+            return d_signal
+
         # Do inplace conversion and set relevant variables.
         if inplace:
             if expanded:
-                for ch in range(self.n_sig):
-                    # NAN locations for the channel
-                    ch_nanlocs = np.isnan(self.e_p_signal[ch])
-                    np.multiply(
-                        self.e_p_signal[ch],
+                for ch, ch_p_signal in enumerate(self.e_p_signal):
+                    ch_d_signal = adc_inplace(
+                        ch_p_signal,
                         self.adc_gain[ch],
-                        self.e_p_signal[ch],
-                    )
-                    np.add(
-                        self.e_p_signal[ch],
                         self.baseline[ch],
-                        self.e_p_signal[ch],
-                    )
-                    np.round(self.e_p_signal[ch], 0, self.e_p_signal[ch])
-                    self.e_p_signal[ch] = self.e_p_signal[ch].astype(
-                        intdtype, copy=False
+                        d_nans[ch],
                     )
-                    self.e_p_signal[ch][ch_nanlocs] = d_nans[ch]
+                    self.e_p_signal[ch] = ch_d_signal
                 self.e_d_signal = self.e_p_signal
                 self.e_p_signal = None
             else:
-                nanlocs = np.isnan(self.p_signal)
-                np.multiply(self.p_signal, self.adc_gain, self.p_signal)
-                np.add(self.p_signal, self.baseline, self.p_signal)
-                np.round(self.p_signal, 0, self.p_signal)
-                self.p_signal = self.p_signal.astype(intdtype, copy=False)
-                self.d_signal = self.p_signal
+                self.d_signal = adc_inplace(
+                    self.p_signal,
+                    self.adc_gain,
+                    self.baseline,
+                    d_nans,
+                )
                 self.p_signal = None
 
         # Return the variable
         else:
             if expanded:
-                d_signal = []
-                for ch in range(self.n_sig):
-                    # NAN locations for the channel
-                    ch_nanlocs = np.isnan(self.e_p_signal[ch])
-                    ch_d_signal = self.e_p_signal[ch].copy()
-                    np.multiply(ch_d_signal, self.adc_gain[ch], ch_d_signal)
-                    np.add(ch_d_signal, self.baseline[ch], ch_d_signal)
-                    np.round(ch_d_signal, 0, ch_d_signal)
-                    ch_d_signal = ch_d_signal.astype(intdtype, copy=False)
-                    ch_d_signal[ch_nanlocs] = d_nans[ch]
-                    d_signal.append(ch_d_signal)
+                e_d_signal = []
+                for ch, ch_p_signal in enumerate(self.e_p_signal):
+                    ch_d_signal = adc_inplace(
+                        ch_p_signal.copy(),
+                        self.adc_gain[ch],
+                        self.baseline[ch],
+                        d_nans[ch],
+                    )
+                    e_d_signal.append(ch_d_signal)
+                return e_d_signal
 
             else:
-                nanlocs = np.isnan(self.p_signal)
-                # Cannot cast dtype to int now because gain is float.
-                d_signal = self.p_signal.copy()
-                np.multiply(d_signal, self.adc_gain, d_signal)
-                np.add(d_signal, self.baseline, d_signal)
-                np.round(d_signal, 0, d_signal)
-                d_signal = d_signal.astype(intdtype, copy=False)
-
-                if nanlocs.any():
-                    for ch in range(d_signal.shape[1]):
-                        if nanlocs[:, ch].any():
-                            d_signal[nanlocs[:, ch], ch] = d_nans[ch]
-
-            return d_signal
+                return adc_inplace(
+                    self.p_signal.copy(),
+                    self.adc_gain,
+                    self.baseline,
+                    d_nans,
+                )
 
     def dac(self, expanded=False, return_res=64, inplace=False):
         """