我们从Python开源项目中，提取了以下50个代码示例，用于说明如何使用numpy.floor()。
def stft(sig, frameSize, overlapFac=0.75, window=np.hanning): """ short time fourier transform of audio signal """ win = window(frameSize) hopSize = int(frameSize - np.floor(overlapFac * frameSize)) # zeros at beginning (thus center of 1st window should be for sample nr. 0) # samples = np.append(np.zeros(np.floor(frameSize / 2.0)), sig) samples = np.array(sig, dtype='float64') # cols for windowing cols = np.ceil((len(samples) - frameSize) / float(hopSize)) + 1 # zeros at end (thus samples can be fully covered by frames) # samples = np.append(samples, np.zeros(frameSize)) frames = stride_tricks.as_strided( samples, shape=(cols, frameSize), strides=(samples.strides[0] * hopSize, samples.strides[0])).copy() frames *= win return np.fft.rfft(frames) # all the definition of the flowing variable can be found # train_net.py
def fftfilt(b, x, *n): N_x = len(x) N_b = len(b) N = 2**np.arange(np.ceil(np.log2(N_b)),np.floor(np.log2(N_x))) cost = np.ceil(N_x / (N - N_b + 1)) * N * (np.log2(N) + 1) N_fft = int(N[np.argmin(cost)]) N_fft = int(N_fft) # Compute the block length: L = int(N_fft - N_b + 1) # Compute the transform of the filter: H = np.fft.fft(b,N_fft) y = np.zeros(N_x, x.dtype) i = 0 while i <= N_x: il = np.min([i+L,N_x]) k = np.min([i+N_fft,N_x]) yt = np.fft.ifft(np.fft.fft(x[i:il],N_fft)*H,N_fft) # Overlap.. y[i:k] = y[i:k] + yt[:k-i] # and add i += L return y
def __call__(self, batch): images, labels = zip(*batch) imgH = self.imgH imgW = self.imgW if self.keep_ratio: ratios = [] for image in images: w, h = image.size ratios.append(w / float(h)) ratios.sort() max_ratio = ratios[-1] imgW = int(np.floor(max_ratio * imgH)) imgW = max(imgH * self.min_ratio, imgW) # assure imgH >= imgW transform = resizeNormalize((imgW, imgH)) images = [transform(image) for image in images] images = torch.cat([t.unsqueeze(0) for t in images], 0) return images, labels
def split_episodes(self, episode_paths, n_train, n_valid, n_test, seed=None, use_all=True): """Split episodes between training, validation and test sets. seed: random seed (have split performed consistently every time)""" if seed is not None: random_state = np.random.get_state() np.random.seed(seed) np.random.shuffle(episode_paths) np.random.set_state(random_state) else: np.random.shuffle(episode_paths) if use_all: multiplier = float(len(episode_paths)) / float(n_train + n_valid + n_test) n_train = int(math.floor(multiplier * n_train)) n_valid = int(math.floor(multiplier * n_valid)) n_test = int(math.floor(multiplier * n_test)) assert n_train + n_valid + n_test <= len(episode_paths) return (episode_paths[:n_train], episode_paths[n_train:n_train + n_valid], episode_paths[n_train + n_test:n_train + n_test + n_test])
def earn_dividend(self, dividend): """ Register the number of shares we held at this dividend's ex date so that we can pay out the correct amount on the dividend's pay date. """ assert dividend['sid'] == self.sid out = {'id': dividend['id']} # stock dividend if dividend['payment_sid']: out['payment_sid'] = dividend['payment_sid'] out['share_count'] = np.floor(self.amount * float(dividend['ratio'])) # cash dividend if dividend['net_amount']: out['cash_amount'] = self.amount * dividend['net_amount'] elif dividend['gross_amount']: out['cash_amount'] = self.amount * dividend['gross_amount'] payment_owed = zp.dividend_payment(out) return payment_owed
def handle_data(self, data): if self.target_shares == 0: assert 0 not in self.portfolio.positions self.order(self.sid(0), 10) self.target_shares = 10 return else: assert self.portfolio.positions[0]['amount'] == \ self.target_shares, "Orders not filled immediately." assert self.portfolio.positions[0]['last_sale_price'] == \ data[0].price, "Orders not filled at current price." self.order_percent(self.sid(0), .001) if isinstance(self.sid(0), Equity): self.target_shares += np.floor( (.001 * self.portfolio.portfolio_value) / data[0].price ) if isinstance(self.sid(0), Future): self.target_shares += np.floor( (.001 * self.portfolio.portfolio_value) / (data[0].price * self.sid(0).multiplier) )
def int_bilin_MT(f, x, y): # assume x, y are in pixel fint = np.zeros(len(x)) for i in range(len(x)): t = y[i] - np.floor(y[i]) u = x[i] - np.floor(x[i]) y0 = f[np.int(np.floor(y[i])), np.int(np.floor(x[i]))] y1 = f[np.int(np.floor(y[i])) + 1, np.int(np.floor(x[i]))] y2 = f[np.int(np.floor(y[i])) + 1, np.int(np.floor(x[i])) + 1] y3 = f[np.int(np.floor(y[i])), np.int(np.floor(x[i])) + 1] fint[i] = t * u * (y0 - y1 + y2 - y3) fint[i] += t * (y1 - y0) fint[i] += u * (y3 - y0) fint[i] += y0 return fint
def expand_to_chunk_size(self, chunk_size, offset=Vec(0,0,0, dtype=int)): """ Align a potentially non-axis aligned bbox to the grid by growing it to the nearest grid lines. Required: chunk_size: arraylike (x,y,z), the size of chunks in the dataset e.g. (64,64,64) Optional: offset: arraylike (x,y,z), the starting coordinate of the dataset """ chunk_size = np.array(chunk_size, dtype=np.float32) result = self.clone() result = result - offset result.minpt = np.floor(result.minpt / chunk_size) * chunk_size result.maxpt = np.ceil(result.maxpt / chunk_size) * chunk_size return result + offset
def shrink_to_chunk_size(self, chunk_size, offset=Vec(0,0,0, dtype=int)): """ Align a potentially non-axis aligned bbox to the grid by shrinking it to the nearest grid lines. Required: chunk_size: arraylike (x,y,z), the size of chunks in the dataset e.g. (64,64,64) Optional: offset: arraylike (x,y,z), the starting coordinate of the dataset """ chunk_size = np.array(chunk_size, dtype=np.float32) result = self.clone() result = result - offset result.minpt = np.ceil(result.minpt / chunk_size) * chunk_size result.maxpt = np.floor(result.maxpt / chunk_size) * chunk_size return result + offset
def resize_image(image,target_shape, pad_value = 0): assert isinstance(target_shape, list) or isinstance(target_shape, tuple) add_shape, subs_shape = [], [] image_shape = image.shape shape_difference = np.asarray(target_shape, dtype=int) - np.asarray(image_shape,dtype=int) for diff in shape_difference: if diff < 0: subs_shape.append(np.s_[int(np.abs(np.ceil(diff/2))):int(np.floor(diff/2))]) add_shape.append((0, 0)) else: subs_shape.append(np.s_[:]) add_shape.append((int(np.ceil(1.0*diff/2)),int(np.floor(1.0*diff/2)))) output = np.pad(image, tuple(add_shape), 'constant', constant_values=(pad_value, pad_value)) output = output[subs_shape] return output
def get_mask_boundaries(self,image_shape,mask_shape,ROI_mask): half_segment_dimensions = np.zeros((len(image_shape), 2), dtype='int32') for index, dim in enumerate(image_shape): if dim % 2 == 0: half_segment_dimensions[index, :] = [dim / 2 - 1, dim / 2] else: half_segment_dimensions[index, :] = [np.floor(dim / 2)] * 2 mask_boundaries = np.zeros(mask_shape, dtype='int32') mask_boundaries[half_segment_dimensions[0][0]:-half_segment_dimensions[0][1], half_segment_dimensions[1][0]:-half_segment_dimensions[1][1], half_segment_dimensions[2][0]:-half_segment_dimensions[2][1]] = 1 if ROI_mask is None: return mask_boundaries else: return mask_boundaries * ROI_mask
def logTickValues(self, minVal, maxVal, size, stdTicks): ## start with the tick spacing given by tickValues(). ## Any level whose spacing is < 1 needs to be converted to log scale ticks = [] for (spacing, t) in stdTicks: if spacing >= 1.0: ticks.append((spacing, t)) if len(ticks) < 3: v1 = int(np.floor(minVal)) v2 = int(np.ceil(maxVal)) #major = list(range(v1+1, v2)) minor = [] for v in range(v1, v2): minor.extend(v + np.log10(np.arange(1, 10))) minor = [x for x in minor if x>minVal and x<maxVal] ticks.append((None, minor)) return ticks
def get_batch_idx(self, idx, **kwargs): if self.mode == 'train': new_idx = [] # self.log.info('Label IDX: {}'.format(idx)) if self.stats_provider is None: label_ids = [ii % self._real_size for ii in idx] else: # print idx, self.stats_provider.get_size() stats_batch = self.stats_provider.get_batch_idx(idx) label_ids = [] for ii in xrange(len(idx)): label_ids.append(np.argmax(stats_batch['y_gt'][ii])) for ii in label_ids: data_group = self.data_provider.label_idx[ii] num_ids = len(data_group) kk = int(np.floor(self.rnd.uniform(0, num_ids))) new_idx.append(data_group[kk]) else: new_idx = idx return self.data_provider.get_batch_idx(new_idx)
def transform(self, images): if self._aug_flag: transformed_images =\ np.zeros([images.shape[0], self._imsize, self._imsize, 3]) ori_size = images.shape[1] for i in range(images.shape[0]): h1 = np.floor((ori_size - self._imsize) * np.random.random()) w1 = np.floor((ori_size - self._imsize) * np.random.random()) cropped_image =\ images[i][w1: w1 + self._imsize, h1: h1 + self._imsize, :] if random.random() > 0.5: transformed_images[i] = np.fliplr(cropped_image) else: transformed_images[i] = cropped_image return transformed_images else: return images
def gaussian_kernel(kernel_shape, sigma=None): """ Get 2D Gaussian kernel :param kernel_shape: kernel size :param sigma: sigma of Gaussian distribution :return: 2D Gaussian kernel """ kern = numpy.zeros((kernel_shape, kernel_shape), dtype='float32') # get sigma from kernel size if sigma is None: sigma = 0.3*((kernel_shape-1.)*0.5 - 1.) + 0.8 def gauss(x, y, s): Z = 2. * numpy.pi * s ** 2. return 1. / Z * numpy.exp(-(x ** 2. + y ** 2.) / (2. * s ** 2.)) mid = numpy.floor(kernel_shape / 2.) for i in xrange(0, kernel_shape): for j in xrange(0, kernel_shape): kern[i, j] = gauss(i - mid, j - mid, sigma) return kern / kern.sum()
def _hpd_interval(self, x, width): """ Code adapted from pymc3.stats.calc_min_interval: https://github.com/pymc-devs/pymc3/blob/master/pymc3/stats.py """ x = np.sort(x) n = len(x) interval_idx_inc = int(np.floor(width * n)) n_intervals = n - interval_idx_inc interval_width = x[interval_idx_inc:] - x[:n_intervals] if len(interval_width) == 0: raise ValueError('Too few elements for interval calculation') min_idx = np.argmin(interval_width) hdi_min = x[min_idx] hdi_max = x[min_idx + interval_idx_inc] index = ['hpd{}_{}'.format(width, x) for x in ['lower', 'upper']] return pd.Series([hdi_min, hdi_max], index=index)
def SLdshear(inputArray, k, axis): """ Computes the discretized shearing operator for a given inputArray, shear number k and axis. This version is adapted such that the MATLAB indexing can be used here in the Python version. """ axis = axis - 1 if k==0: return inputArray rows = np.asarray(inputArray.shape)[0] cols = np.asarray(inputArray.shape)[1] shearedArray = np.zeros((rows, cols), dtype=inputArray.dtype) if axis == 0: for col in range(cols): shearedArray[:,col] = np.roll(inputArray[:,col], int(k * np.floor(cols/2-col))) else: for row in range(rows): shearedArray[row,:] = np.roll(inputArray[row,:], int(k * np.floor(rows/2-row))) return shearedArray
def value_to_bin_index(val, **kwargs): """Convert value to bin index Convert a numeric or timestamp column to an integer bin index. :param bin_width: bin_width value needed to convert column to an integer bin index :param bin_offset: bin_offset value needed to convert column to an integer bin index """ try: # NOTE this notation also works for timestamps bin_width = kwargs.get('bin_width', 1) bin_offset = kwargs.get('bin_offset', 0) bin_index = int(np.floor((val - bin_offset) / bin_width)) return bin_index except BaseException: pass return val
def value_to_bin_center(val, **kwargs): """Convert value to bin center Convert a numeric or timestamp column to a common bin center value. :param bin_width: bin_width value needed to convert column to a common bin center value :param bin_offset: bin_offset value needed to convert column to a common bin center value """ try: # NOTE this notation also works for timestamps, and does not change the # unit bin_width = kwargs.get('bin_width', 1) bin_offset = kwargs.get('bin_offset', 0) bin_index = int(np.floor((val - bin_offset) / bin_width)) obj_type = type(bin_width) return bin_offset + obj_type((bin_index + 0.5) * bin_width) except BaseException: pass return val
def save_fft(fil,audio_in): samples = len(audio_in) fft_size = 2**int(floor(log(samples)/log(2.0))) freq = fft(audio_in[0:fft_size]) s_data = numpy.zeros(fft_size/2) x_data = numpy.zeros(fft_size/2) peak = 0; for j in xrange(fft_size/2): if (abs(freq[j]) > peak): peak = abs(freq[j]) for j in xrange(fft_size/2): x_data[j] = log(2.0*(j+1.0)/fft_size); if (x_data[j] < -10): x_data[j] = -10 s_data[j] = 10.0*log(abs(freq[j])/peak)/log(10.0) plt.ylim([-50,0]) plt.plot(x_data,s_data) plt.title('fft log power') plt.grid() fields = fil.split('.') plt.savefig(fields[0]+'_fft.png', bbox_inches="tight") plt.clf() plt.close()
def _gene_embed_space(self,vec): shape = vec.shape vec = vec.flatten() combo_neg_idx = np.array([1 if vec[i]<0 else 0 for i in range(len(vec))]) vec_pos = np.abs(vec) int_part = np.floor(vec_pos) frac_part = np.round(vec_pos - int_part,2) bi_int_part=[] #?????????????signature??????? for i in range(len(int_part)): bi=list(bin(int(int_part[i]))[2:]) bie = [0] * (16 - len(bi)) bie.extend(bi) bi_int_part.append(np.array(bie,dtype=np.uint16)) bi_int_part = np.array(bi_int_part) sig = [] for i in range(len(bi_int_part)): sig.append(bi_int_part[i][10]) sig = np.array(sig).reshape(shape) return np.array(bi_int_part),frac_part.reshape(shape),combo_neg_idx.reshape(shape),sig
def M(self): """Returns the :math:`M` matrix of integers that determine points at which the functions are sampled in the unit cell. Examples: For `S = [2, 2, 1]`, the returned matrix is: .. code-block:: python np.ndarray([[0,0,0], [1,0,0], [0,1,0], [1,1,0]], dtype=int) """ if self._M is None: ms = np.arange(np.prod(self.S, dtype=int)) m1 = np.fmod(ms, self.S[0]) m2 = np.fmod(np.floor(ms/self.S[0]), self.S[1]) m3 = np.fmod(np.floor(ms/(self.S[0]*self.S[1])), self.S[2]) #Make sure we explicitly use an integer array; it's faster. self._M = np.asarray(np.vstack((m1, m2, m3)).T, dtype=int) return self._M
def _latvec_plot(self, R=True, withpts=False, legend=False): """Plots the lattice vectors (for real or reciprocal space). """ import matplotlib.pyplot as plt from mpl_toolkits.mplot3d import Axes3D fig = plt.figure() ax = fig.gca(projection='3d') vecs = self.R if R else self.K for i in range(3): steps = np.linspace(0, 1, np.floor(10*np.linalg.norm(vecs[:,i]))) Ri = vecs[:,i] Ri.shape = (1, 3) steps.shape = (len(steps), 1) line = np.dot(steps, Ri) ax.plot(line[:,0], line[:,1], line[:,2], label="R{0:d}".format(i+1)) if withpts: pts = self.r if R else self.G ax.scatter(pts[:,0], pts[:,1], pts[:,2], color='k') if legend: ax.legend() return (fig, ax)
def cumultativesumstest(binin): ''' The focus of this test is the maximal excursion (from zero) of the random walk defined by the cumulative sum of adjusted (-1, +1) digits in the sequence. The purpose of the test is to determine whether the cumulative sum of the partial sequences occurring in the tested sequence is too large or too small relative to the expected behavior of that cumulative sum for random sequences. This cumulative sum may be considered as a random walk. For a random sequence, the random walk should be near zero. For non-random sequences, the excursions of this random walk away from zero will be too large.''' n = len(binin) ss = [int(el) for el in binin] sc = map(sumi, ss) cs = np.cumsum(sc) z = max(abs(cs)) ra = 0 start = int(np.floor(0.25 * np.floor(-n / z) + 1)) stop = int(np.floor(0.25 * np.floor(n / z) - 1)) pv1 = [] for k in xrange(start, stop + 1): pv1.append(sst.norm.cdf((4 * k + 1) * z / np.sqrt(n)) - sst.norm.cdf((4 * k - 1) * z / np.sqrt(n))) start = int(np.floor(0.25 * np.floor(-n / z - 3))) stop = int(np.floor(0.25 * np.floor(n / z) - 1)) pv2 = [] for k in xrange(start, stop + 1): pv2.append(sst.norm.cdf((4 * k + 3) * z / np.sqrt(n)) - sst.norm.cdf((4 * k + 1) * z / np.sqrt(n))) pval = 1 pval -= reduce(su, pv1) pval += reduce(su, pv2) return pval
def rel_crop(im, rel_cx, rel_cy, crop_size): map_size = im.shape[1] r = crop_size / 2 abs_cx = rel_cx * map_size abs_cy = rel_cy * map_size na = np.floor([abs_cy-r, abs_cy+r, abs_cx-r, abs_cx+r]).astype(np.int32) a = np.clip(na, 0, map_size) px0 = a[2] - na[2] px1 = na[3] - a[3] py0 = a[0] - na[0] py1 = na[1] - a[1] crop = im[a[0]:a[1], a[2]:a[3]] crop = np.pad(crop, ((py0, py1), (px0, px1), (0, 0)), mode='reflect') assert crop.shape == (crop_size, crop_size, im.shape[2]) return crop
def _downsample_mask(X, pct): """ Create a boolean mask indicating which subset of X should be evaluated. """ if pct < 1.0: Mask = np.zeros(X.shape, dtype=np.bool) m = X.shape[-2] n = X.shape[-1] nToEval = np.round(pct*m*n).astype(np.int32) idx = sobol(2, nToEval ,0) idx[0] = np.floor(m*idx[0]) idx[1] = np.floor(n*idx[1]) idx = idx.astype(np.int32) Mask[:,:,idx[0], idx[1]] = True else: Mask = np.ones(X.shape, dtype=np.bool) return Mask
def update_output(self, x): N, C, H, W = x.shape pool_height, pool_width = self.kW, self.kH stride = self.dW assert ( H - pool_height) % stride == 0 or H == pool_height, 'Invalid height' assert ( W - pool_width) % stride == 0 or W == pool_width, 'Invalid width' out_height = int(np.floor((H - pool_height) / stride + 1)) out_width = int(np.floor((W - pool_width) / stride + 1)) x_split = x.reshape(N * C, 1, H, W) x_cols = im2col_cython( x_split, pool_height, pool_width, padding=0, stride=stride) x_cols_avg = np.mean(x_cols, axis=0) out = x_cols_avg.reshape( out_height, out_width, N, C).transpose(2, 3, 0, 1) self.x_shape = x.shape self.x_cols = x_cols self.output = out return self.output
def extract_top_plane_nodes(nodefile, top_face): """ :param nodefile: :param top_face: :return: planeNodeIDs """ import numpy as np import fem_mesh top_face = np.array(top_face) nodeIDcoords = fem_mesh.load_nodeIDs_coords(nodefile) [snic, axes] = fem_mesh.SortNodeIDs(nodeIDcoords) # extract spatially-sorted node IDs on a the top z plane axis = int(np.floor(np.divide(top_face.nonzero(), 2))) if np.mod(top_face.nonzero(), 2) == 1: plane = (axis, axes[axis].max()) else: plane = (axis, axes[axis].min()) planeNodeIDs = fem_mesh.extractPlane(snic, axes, plane) return planeNodeIDs
def timer(s, v='', nloop=500, nrep=3): units = ["s", "ms", "µs", "ns"] scaling = [1, 1e3, 1e6, 1e9] print("%s : %-50s : " % (v, s), end=' ') varnames = ["%ss,nm%ss,%sl,nm%sl" % tuple(x*4) for x in 'xyz'] setup = 'from __main__ import numpy, ma, %s' % ','.join(varnames) Timer = timeit.Timer(stmt=s, setup=setup) best = min(Timer.repeat(nrep, nloop)) / nloop if best > 0.0: order = min(-int(numpy.floor(numpy.log10(best)) // 3), 3) else: order = 3 print("%d loops, best of %d: %.*g %s per loop" % (nloop, nrep, 3, best * scaling[order], units[order]))
def dec_round(num, dprec=4, rnd='down', rto_zero=False): """ Round up/down numeric ``num`` at specified decimal ``dprec``. Parameters ---------- num: float dprec: int Decimal position for truncation. rnd: str (default: 'down') Set as 'up' or 'down' to return a rounded-up or rounded-down value. rto_zero: bool (default: False) Use a *round-towards-zero* method, e.g., ``floor(-3.5) == -3``. Returns ---------- float (default: rounded-up) """ dprec = 10**dprec if rnd == 'up' or (rnd == 'down' and rto_zero and num < 0.): return np.ceil(num*dprec)/dprec elif rnd == 'down' or (rnd == 'up' and rto_zero and num < 0.): return np.floor(num*dprec)/dprec return np.round(num, dprec)
def shuffle_to_training_data(self, expert_data, on_policy_data, expert_fail_data): data = np.vstack([expert_data['data'], on_policy_data['data'], expert_fail_data['data']]) classes = np.vstack([expert_data['classes'], on_policy_data['classes'], expert_fail_data['classes']]) domains = np.vstack([expert_data['domains'], on_policy_data['domains'], expert_fail_data['domains']]) sample_range = data.shape[0]*data.shape[1] all_idxs = np.random.permutation(sample_range) t_steps = data.shape[1] data_matrix = np.zeros(shape=(sample_range, self.im_height, self.im_width, self.im_channels)) data_matrix_two = np.zeros(shape=(sample_range, self.im_height, self.im_width, self.im_channels)) class_matrix = np.zeros(shape=(sample_range, 2)) dom_matrix = np.zeros(shape=(sample_range, 2)) for one_idx, iter_step in zip(all_idxs, range(0, sample_range)): traj_key = np.floor(one_idx/t_steps) time_key = one_idx % t_steps time_key_plus_one = min(time_key + 3, t_steps-1) data_matrix[iter_step, :, :, :] = data[traj_key, time_key, :, :, :] data_matrix_two[iter_step, :, :, :] = data[traj_key, time_key_plus_one, :, :, :] class_matrix[iter_step, :] = classes[traj_key, time_key, :] dom_matrix[iter_step, :] = domains[traj_key, time_key, :] return data_matrix, data_matrix_two, dom_matrix, class_matrix
def transform_to_yolo_labels(self, labels): """ Transform voc_label_parser' result to yolo label. :param labels: [is_obj, x, y, w, h, class_probs..], ... :return: yolo label """ yolo_label = np.zeros([self.side, self.side, (1 + self.coords) + self.classes]).astype(np.float32) shuffle(labels) for label in labels: yolo_box = self.convert_to_yolo_box(self.ori_im_shape[::-1], list(label[2:])) assert np.max(yolo_box) < 1 [loc_y, loc_x] = [int(np.floor(yolo_box[1] * self.side)), int(np.floor(yolo_box[0] * self.side))] yolo_label[loc_y][loc_x][0] = 1.0 # is obj yolo_label[loc_y][loc_x][1:5] = yolo_box # bbox yolo_label[loc_y][loc_x][5:] = 0 # only one obj in one grid yolo_label[loc_y][loc_x][4+label[0]] = 1.0 # class return yolo_label
def julian_date(hUTC, dayofyear, year): """ Julian calendar date Args: hUTC: fractional hour (UTC time) dayofyear (int): year (int): Returns: the julian date Details: World Meteorological Organization (2006).Guide to meteorological instruments and methods of observation. Geneva, Switzerland. """ delta = year - 1949 leap = numpy.floor(delta / 4.) return 2432916.5 + delta * 365 + leap + dayofyear + hUTC / 24.
def unknown_feature_extractor(x, sr, win_len, shift_len, barks, inner_win, inner_shift, win_type, method_version): x_spectrum = stft_extractor(x, win_len, shift_len, win_type) coef = get_fft_bark_mat(sr, win_len, barks, 20, sr//2) bark_spect = np.matmul(coef, x_spectrum) ams = np.zeros((barks, inner_win//2+1, (bark_spect.shape[1] - inner_win)//inner_shift)) for i in range(barks): channel_stft = stft_extractor(bark_spect[i, :], inner_win, inner_shift, 'hanning') if method_version == 'v1': ams[i, :, :] = 20 * np.log(np.abs(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift])) elif method_version == 'v2': channel_amplitude = np.abs(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift]) channel_angle = np.angle(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift]) channel_angle = channel_angle - (np.floor(channel_angle / (2.*np.pi)) * (2.*np.pi)) ams[i, :, :] = np.power(channel_amplitude, 1./3.) * channel_angle else: ams[i, :, :] = np.abs(channel_stft) return ams
def ams_extractor(x, sr, win_len, shift_len, barks, inner_win, inner_shift, win_type, method_version): x_spectrum = stft_extractor(x, win_len, shift_len, win_type) coef = get_fft_bark_mat(sr, win_len, barks, 20, sr//2) bark_spect = np.matmul(coef, x_spectrum) ams = np.zeros((barks, inner_win//2+1, (bark_spect.shape[1] - inner_win)//inner_shift)) for i in range(barks): channel_stft = stft_extractor(bark_spect[i, :], inner_win, inner_shift, 'hanning') if method_version == 'v1': ams[i, :, :] = 20 * np.log(np.abs(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift])) elif method_version == 'v2': channel_amplitude = np.abs(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift]) channel_angle = np.angle(channel_stft[:inner_win//2+1, :(bark_spect.shape[1] - inner_win)//inner_shift]) channel_angle = channel_angle - (np.floor(channel_angle / (2.*np.pi)) * (2.*np.pi)) ams[i, :, :] = np.power(channel_amplitude, 1./3.) * channel_angle else: ams[i, :, :] = np.abs(channel_stft) return ams
def getTimeDerivative(I, Win): dw = np.floor(Win/2) t = np.arange(-dw, dw+1) sigma = 0.4*dw xgaussf = t*np.exp(-t**2 / (2*sigma**2)) #Normalize by L1 norm to control for length of window xgaussf = xgaussf/np.sum(np.abs(xgaussf)) xgaussf = xgaussf[:, None] IRet = scipy.signal.convolve2d(I, xgaussf, 'valid') validIdx = np.arange(dw, I.shape[0]-dw, dtype='int64') return [IRet, validIdx] ############################################################# #### FAST TIME DELAY EMBEDDING, Tau = 1 ##### ############################################################# #Input: I: P x N Video with frames along the columns #W: Windows #Ouput: Mu: P x W video with mean frames along the columns
def mu_law(x, mu=255, int8=False): """A TF implementation of Mu-Law encoding. Args: x: The audio samples to encode. mu: The Mu to use in our Mu-Law. int8: Use int8 encoding. Returns: out: The Mu-Law encoded int8 data. """ out = tf.sign(x) * tf.log(1 + mu * tf.abs(x)) / np.log(1 + mu) out = tf.floor(out * 128) if int8: out = tf.cast(out, tf.int8) return out
def impad_gpu(y_gpu, sf): sf = np.array(sf) shape = (np.array(y_gpu.shape) + sf).astype(np.uint32) dtype = y_gpu.dtype block_size = (16,16,1) grid_size = (int(np.ceil(float(shape[1])/block_size[0])), int(np.ceil(float(shape[0])/block_size[1]))) preproc = _generate_preproc(dtype, shape) mod = SourceModule(preproc + kernel_code, keep=True) padded_gpu = cua.empty((int(shape[0]), int(shape[1])), dtype) impad_fun = mod.get_function("impad") upper_left = np.uint32(np.floor(sf / 2.)) original_size = np.uint32(np.array(y_gpu.shape)) impad_fun(padded_gpu.gpudata, y_gpu.gpudata, upper_left[1], upper_left[0], original_size[0], original_size[1], block=block_size, grid=grid_size) return padded_gpu
def laplace_stack_gpu(y_gpu, mode='valid'): """ This funtion computes the Laplacian of each slice of a stack of images """ shape = np.array(y_gpu.shape).astype(np.uint32) dtype = y_gpu.dtype block_size = (6,int(np.floor(512./6./float(shape[0]))),int(shape[0])) grid_size = (int(np.ceil(float(shape[1])/block_size[0])), int(np.ceil(float(shape[0])/block_size[1]))) shared_size = int((2+block_size[0])*(2+block_size[1])*(2+block_size[2]) *dtype.itemsize) preproc = _generate_preproc(dtype, (shape[1],shape[2])) mod = SourceModule(preproc + kernel_code, keep=True) laplace_fun_gpu = mod.get_function("laplace_stack_same") laplace_gpu = cua.empty((y_gpu.shape[0], y_gpu.shape[1], y_gpu.shape[2]), y_gpu.dtype) laplace_fun_gpu(laplace_gpu.gpudata, y_gpu.gpudata, block=block_size, grid=grid_size, shared=shared_size) return laplace_gpu
def laplace3d_gpu(y_gpu): shape = np.array(y_gpu.shape).astype(np.uint32) dtype = y_gpu.dtype block_size = (6,int(np.floor(512./6./float(shape[0]))),int(shape[0])) grid_size = (int(np.ceil(float(shape[1])/block_size[0])), int(np.ceil(float(shape[0])/block_size[1]))) shared_size = int((2+block_size[0])*(2+block_size[1])*(2+block_size[2]) *dtype.itemsize) preproc = _generate_preproc(dtype, (shape[1],shape[2])) mod = SourceModule(preproc + kernel_code, keep=True) laplace_fun_gpu = mod.get_function("laplace3d_same") laplace_gpu = cua.empty((y_gpu.shape[0], y_gpu.shape[1], y_gpu.shape[2]), y_gpu.dtype) laplace_fun_gpu(laplace_gpu.gpudata, y_gpu.gpudata, block=block_size, grid=grid_size, shared=shared_size) return laplace_gpu
def wsparsify(w_gpu, percentage): """ Keeps only as many entries nonzero as specified by percentage. """ w = w_gpu.get() vals = sort(w)[::-1] idx = floor(prod(w.shape()) * percentage/100) zw_gpu = cua.zeros_like(w_gpu) # gpu array filled with zeros tw_gpu = cua.empty_like(w_gpu) # gpu array containing threshold tw_gpu.fill(vals[idx]) w_gpu = cua.if_positive(w_gpu > tw_gpu, w_gpu, zw_gpu) del zw_gpu del tw_gpu return w_gpu
def sparsify(x, percentage): """ Keeps only as many entries nonzero as specified by percentage. Note that only the larges values are kept. -------------------------------------------------------------------------- Usage: Call: y = sparsify(x, percentage) Input: x input ndarray x percentage percentage of nonzero entries in y Output: sparsified version of x -------------------------------------------------------------------------- Copyright (C) 2011 Michael Hirsch """ vals = np.sort(x.flatten())[::-1] idx = np.floor(np.prod(x.shape) * percentage/100) x[x < vals[idx]] = 0 return x
def stft(sig, frameSize, overlapFac=0.75, window=np.hanning): """ short time fourier transform of audio signal """ win = window(frameSize) hopSize = int(frameSize - np.floor(overlapFac * frameSize)) # zeros at beginning (thus center of 1st window should be for sample nr. 0) # samples = np.append(np.zeros(np.floor(frameSize / 2.0)), sig) samples = np.array(sig, dtype='float64') # cols for windowing cols = np.floor((len(samples) - frameSize) / float(hopSize)) # zeros at end (thus samples can be fully covered by frames) # samples = np.append(samples, np.zeros(frameSize)) frames = stride_tricks.as_strided( samples, shape=(cols, frameSize), strides=(samples.strides[0] * hopSize, samples.strides[0])).copy() frames *= win return np.fft.rfft(frames)
def stft(sig, frameSize, overlapFac=0.75, window=np.hanning): """ short time fourier transform of audio signal """ win = window(frameSize) hopSize = int(frameSize - np.floor(overlapFac * frameSize)) # zeros at beginning (thus center of 1st window should be for sample nr. 0) # samples = np.append(np.zeros(np.floor(frameSize / 2.0)), sig) samples = np.array(sig, dtype='float64') # cols for windowing cols = np.ceil((len(samples) - frameSize) / float(hopSize)) + 1 # zeros at end (thus samples can be fully covered by frames) # samples = np.append(samples, np.zeros(frameSize)) frames = stride_tricks.as_strided( samples, shape=(cols, frameSize), strides=(samples.strides[0] * hopSize, samples.strides[0])).copy() frames *= win return np.fft.rfft(frames)
