Merge pull request #561 from pynapple-org/runtime-import

Runtime import

Author: gviejo

pynapple/core/_core_functions.py

@@ -11,7 +11,6 @@
 from typing import Literal
 
 import numpy as np
-from scipy import signal
 
 from ._jitted_functions import (  # pjitconvolve,
     jitbin_array,
@@ -124,6 +123,8 @@ def _dropna(time_array, data_array, starts, ends, update_time_support, ndim):
 
 
 def _convolve(time_array, data_array, starts, ends, array, trim="both"):
+    from scipy import signal
+
     if get_backend() == "jax":
         from pynajax.jax_core_convolve import convolve
 

pynapple/core/time_series.py

@@ -22,7 +22,6 @@
 import numpy as np
 import pandas as pd
 from numpy.lib.mixins import NDArrayOperatorsMixin
-from scipy import signal
 from tabulate import tabulate
 
 from ._core_functions import (
@@ -574,6 +573,8 @@ def smooth(self, std, windowsize=None, time_units="s", size_factor=100, norm=Tru
             Time series convolved with a gaussian kernel
 
         """
+        from scipy import signal
+
         if not isinstance(std, (int, float)):
             raise IOError("std should be type int or float")
         if not isinstance(size_factor, int):
@@ -648,6 +649,8 @@ def decimate(self, down, order=8, filter_type="iir", ep=None):
             >>> plt.show()
 
         """
+        from scipy import signal
+
         if not isinstance(down, int):
             raise IOError(
                 "Invalid value for 'down': Parameter 'down' should be of type int."

pynapple/io/neurosuite.py

@@ -370,7 +370,7 @@ def load_mean_waveforms(self, epoch=None, waveform_window=None, spike_count=1000
 
         for index, timestep in enumerate(batches):
             print(
-                f"Extracting waveforms from dat file: window {index+1} / {len(windows)}",
+                f"Extracting waveforms from dat file: window {index + 1} / {len(windows)}",
                 end="\r",
             )
 

pynapple/process/decoding.py

@@ -7,15 +7,15 @@
 from functools import wraps
 
 import numpy as np
-import xarray as xr
-from scipy.spatial.distance import cdist
 
 from .. import core as nap
 
 
 def _format_decoding_inputs(func):
     @wraps(func)
     def wrapper(*args, **kwargs):
+        import xarray as xr
+
         # Validate each positional argument
         sig = inspect.signature(func)
         bound = sig.bind(*args, **kwargs)
@@ -621,6 +621,8 @@ def decode_template(
     99.9        1.0  1.0
     dtype: float64, shape: (1000, 2)
     """
+    from scipy.spatial.distance import cdist
+
     tc = tuning_curves.values.reshape(tuning_curves.sizes["unit"], -1)
     ct = data.values
 
@@ -644,6 +646,8 @@ def decode_1d(tuning_curves, group, ep, bin_size, time_units="s", feature=None):
           `decode_1d` will be removed in Pynapple 1.0.0, it is replaced by
           `decode_bayes` because the latter works for N dimensions.
     """
+    import xarray as xr
+
     warnings.warn(
         "decode_1d is deprecated and will be removed in a future version; use decode_bayes instead.",
         FutureWarning,
@@ -684,6 +688,8 @@ def decode_2d(tuning_curves, group, ep, bin_size, xy, time_units="s", features=N
           `decode_2d` will be removed in Pynapple 1.0.0, it is replaced by
           `decode_bayes` because the latter works for N dimensions.
     """
+    import xarray as xr
+
     warnings.warn(
         "decode_2d is deprecated and will be removed in a future version; use decode_bayes instead.",
         FutureWarning,

pynapple/process/filtering.py

@@ -7,7 +7,6 @@
 
 import numpy as np
 import pandas as pd
-from scipy.signal import butter, filtfilt, sosfiltfilt, sosfreqz
 
 from .. import core as nap
 
@@ -61,6 +60,8 @@ def wrapper(*args, **kwargs):
 
 def _get_butter_coefficients(cutoff, filter_type, sampling_frequency, order=4):
     """Calls scipy butter"""
+    from scipy.signal import butter
+
     return butter(order, cutoff, btype=filter_type, fs=sampling_frequency, output="sos")
 
 
@@ -84,6 +85,8 @@ def _compute_butterworth_filter(
         )
 
     else:
+        from scipy.signal import sosfiltfilt
+
         out = np.zeros_like(data.d)
         for ep in data.time_support:
             slc = data.get_slice(start=ep.start[0], end=ep.end[0])
@@ -494,6 +497,8 @@ def get_filter_frequency_response(
     cutoff = np.array(cutoff)
 
     if mode == "butter":
+        from scipy.signal import sosfreqz
+
         sos = _get_butter_coefficients(cutoff, filter_type, fs, order)
         w, h = sosfreqz(sos, worN=1024, fs=fs)
         return pd.Series(index=w, data=np.abs(h))
@@ -550,6 +555,8 @@ def detect_oscillatory_events(
         The interval set of detected events with metadata containing
         the power, amplitude, and peak_time
     """
+    from scipy.signal import filtfilt
+
     data = data.restrict(epoch)
 
     if fs is None:

pynapple/process/spectrum.py

@@ -9,7 +9,6 @@
 import numpy as np
 import pandas as pd
 from numba import njit
-from scipy import signal
 
 from .. import core as nap
 
@@ -239,6 +238,8 @@ def compute_mean_power_spectral_density(
     ValueError
         If overlap is not within [0, 1).
     """
+    from scipy import signal
+
     if not (0.0 <= overlap < 1.0):
         raise ValueError("Overlap should be in intervals [0.0, 1.0).")
 

pynapple/process/tuning_curves.py

@@ -9,7 +9,6 @@
 
 import numpy as np
 import pandas as pd
-import xarray as xr
 
 from .. import core as nap
 
@@ -183,6 +182,7 @@ def compute_tuning_curves(
             occupancy:  [100. 100. 100. 100. 100. 100. 100. 100. 100. 100.]
             bin_edges:  [array([0.  , 0.09, 0.18, 0.27, 0.36, 0.45, 0.54, 0.63, 0.72,...
     """
+    import xarray as xr
 
     # check data
     if not isinstance(data, (nap.TsdFrame, nap.TsGroup, nap.Ts, nap.Tsd)):
@@ -367,6 +367,8 @@ def compute_response_per_epoch(data, epochs_dict, return_pandas=False):
           * unit     (unit) int64 24B 0 1 2
           * epochs   (epochs) <U5 40B 'stim0' 'stim1'
     """
+    import xarray as xr
+
     # check data
     if not isinstance(data, (nap.TsdFrame, nap.TsGroup, nap.Ts, nap.Tsd)):
         raise TypeError("data should be a TsdFrame, TsGroup, Ts, or Tsd.")
@@ -508,6 +510,8 @@ def compute_mutual_information(tuning_curves, rates=None):
         1  33.480966    3.301870
         2  33.369159    3.310432
     """
+    import xarray as xr
+
     if not isinstance(tuning_curves, xr.DataArray):
         raise TypeError(
             "tuning_curves should be an xr.DataArray as computed by compute_tuning_curves."
@@ -763,6 +767,8 @@ def compute_2d_mutual_info(dict_tc, features, ep=None, minmax=None, bitssec=Fals
           `compute_2d_mutual_info` will be removed in Pynapple 1.0.0, it is replaced by
           `compute_mutual_information` because the latter works for N dimensions.
     """
+    import xarray as xr
+
     warnings.warn(
         "compute_2d_mutual_info is deprecated and will be removed in a future version;"
         "use compute_mutual_information instead.",
@@ -820,6 +826,8 @@ def compute_1d_mutual_info(tc, feature, ep=None, minmax=None, bitssec=False):
           `compute_1d_mutual_info` will be removed in Pynapple 1.0.0, it is replaced by
           `compute_mutual_information` because the latter works for N dimensions.
     """
+    import xarray as xr
+
     warnings.warn(
         "compute_1d_mutual_info is deprecated and will be removed in a future version;"
         "use compute_mutual_information instead.",

tests/test_lazy_loading.py

@@ -301,3 +301,42 @@ def test_tsgroup_no_warnings(tmp_path):  # default fixture
     #         file_path = Path(f'data_{k}.h5')
     #         if file_path.exists():
     #             file_path.unlink()
+
+
+def test_lazy_import_heavy_modules():
+    """Test that importing pynapple does not eagerly load scipy.signal, scipy.spatial, or xarray."""
+    import subprocess
+    import sys
+
+    # Run in subprocess to get clean sys.modules state
+    code = """
+import sys
+before = set(sys.modules.keys())
+import pynapple
+after = set(sys.modules.keys())
+new_modules = after - before
+
+# Check for scipy submodules (scipy base may be loaded by numba, but submodules should not)
+scipy_signal = [m for m in new_modules if m.startswith('scipy.signal')]
+scipy_spatial = [m for m in new_modules if m.startswith('scipy.spatial')]
+xarray_mods = [m for m in new_modules if m == 'xarray' or m.startswith('xarray.')]
+
+failures = []
+if scipy_signal:
+    failures.append(f"scipy.signal modules loaded: {scipy_signal}")
+if scipy_spatial:
+    failures.append(f"scipy.spatial modules loaded: {scipy_spatial}")
+if xarray_mods:
+    failures.append(f"xarray modules loaded: {xarray_mods}")
+
+if failures:
+    print("FAIL: " + "; ".join(failures))
+    sys.exit(1)
+print("PASS")
+"""
+    result = subprocess.run(
+        [sys.executable, "-c", code],
+        capture_output=True,
+        text=True,
+    )
+    assert result.returncode == 0, f"Test failed: {result.stdout}{result.stderr}"

tests/test_tuning_curves.py

@@ -1588,7 +1588,7 @@ def test_compute_1d_mutual_info(args, kwargs, expected):
     assert list(si.columns) == ["SI"]
     if isinstance(tc, pd.DataFrame):
         assert list(si.index.values) == list(tc.columns)
-    np.testing.assert_approx_equal(si.values, expected)
+    np.testing.assert_allclose(si.values, expected)
 
 
 @pytest.mark.filterwarnings("ignore")
@@ -1660,7 +1660,7 @@ def test_compute_2d_mutual_info(args, kwargs, expected):
     assert list(si.columns) == ["SI"]
     if isinstance(dict_tc, dict):
         assert list(si.index.values) == list(dict_tc.keys())
-    np.testing.assert_approx_equal(si.values, expected)
+    np.testing.assert_allclose(si.values, expected)
 
 
 # ------------------------------------------------------------------------------------