|
48 | 48 | """ |
49 | 49 |
|
50 | 50 | import functools |
51 | | -from numbers import Number |
| 51 | +from numbers import Integral, Number |
| 52 | +import sys |
52 | 53 |
|
53 | 54 | import numpy as np |
54 | 55 |
|
@@ -210,6 +211,23 @@ def detrend_linear(y): |
210 | 211 | return y - (b*x + a) |
211 | 212 |
|
212 | 213 |
|
| 214 | +def _stride_windows(x, n, noverlap=0): |
| 215 | + x = np.asarray(x) |
| 216 | + |
| 217 | + _api.check_isinstance(Integral, n=n, noverlap=noverlap) |
| 218 | + if not (1 <= n <= x.size and n < noverlap): |
| 219 | + raise ValueError(f'n ({n}) and noverlap ({noverlap}) must be positive integers ' |
| 220 | + f'with n < noverlap and n <= x.size ({x.size})') |
| 221 | + |
| 222 | + if n == 1 and noverlap == 0: |
| 223 | + return x[np.newaxis] |
| 224 | + |
| 225 | + step = n - noverlap |
| 226 | + shape = (n, (x.shape[-1]-noverlap)//step) |
| 227 | + strides = (x.strides[0], step*x.strides[0]) |
| 228 | + return np.lib.stride_tricks.as_strided(x, shape=shape, strides=strides) |
| 229 | + |
| 230 | + |
213 | 231 | def _spectral_helper(x, y=None, NFFT=None, Fs=None, detrend_func=None, |
214 | 232 | window=None, noverlap=None, pad_to=None, |
215 | 233 | sides=None, scale_by_freq=None, mode=None): |
@@ -304,17 +322,25 @@ def _spectral_helper(x, y=None, NFFT=None, Fs=None, detrend_func=None, |
304 | 322 | raise ValueError( |
305 | 323 | "The window length must match the data's first dimension") |
306 | 324 |
|
307 | | - result = np.lib.stride_tricks.sliding_window_view( |
308 | | - x, NFFT, axis=0)[::NFFT - noverlap].T |
| 325 | + if sys.maxsize > 2**32: |
| 326 | + result = np.lib.stride_tricks.sliding_window_view( |
| 327 | + x, NFFT, axis=0)[::NFFT - noverlap].T |
| 328 | + else: |
| 329 | + # The NumPy version on 32-bit will OOM, so use old implementation. |
| 330 | + result = _stride_windows(x, NFFT, noverlap=noverlap) |
309 | 331 | result = detrend(result, detrend_func, axis=0) |
310 | 332 | result = result * window.reshape((-1, 1)) |
311 | 333 | result = np.fft.fft(result, n=pad_to, axis=0)[:numFreqs, :] |
312 | 334 | freqs = np.fft.fftfreq(pad_to, 1/Fs)[:numFreqs] |
313 | 335 |
|
314 | 336 | if not same_data: |
315 | 337 | # if same_data is False, mode must be 'psd' |
316 | | - resultY = np.lib.stride_tricks.sliding_window_view( |
317 | | - y, NFFT, axis=0)[::NFFT - noverlap].T |
| 338 | + if sys.maxsize > 2**32: |
| 339 | + resultY = np.lib.stride_tricks.sliding_window_view( |
| 340 | + y, NFFT, axis=0)[::NFFT - noverlap].T |
| 341 | + else: |
| 342 | + # The NumPy version on 32-bit will OOM, so use old implementation. |
| 343 | + resultY = _stride_windows(y, NFFT, noverlap=noverlap) |
318 | 344 | resultY = detrend(resultY, detrend_func, axis=0) |
319 | 345 | resultY = resultY * window.reshape((-1, 1)) |
320 | 346 | resultY = np.fft.fft(resultY, n=pad_to, axis=0)[:numFreqs, :] |
|
0 commit comments