mi@0: """
mi@0: Set of util functions for the section similarity project.
mi@0: """
mi@0: 
mi@0: import copy
mi@0: import numpy as np
mi@0: import json
mi@0: import scipy.fftpack
mi@0: import pylab as plt
mi@0: 
mi@0: def resample_mx(X, incolpos, outcolpos):
mi@0:     """
mi@0:     Y = resample_mx(X, incolpos, outcolpos)
mi@0:     X is taken as a set of columns, each starting at 'time'
mi@0:     colpos, and continuing until the start of the next column.
mi@0:     Y is a similar matrix, with time boundaries defined by
mi@0:     outcolpos.  Each column of Y is a duration-weighted average of
mi@0:     the overlapping columns of X.
mi@0:     2010-04-14 Dan Ellis dpwe@ee.columbia.edu  based on samplemx/beatavg
mi@0:     -> python: TBM, 2011-11-05, TESTED
mi@0:     """
mi@0:     noutcols = len(outcolpos)
mi@0:     Y = np.zeros((X.shape[0], noutcols))
mi@0:     # assign 'end times' to final columns
mi@0:     if outcolpos.max() > incolpos.max():
mi@0:         incolpos = np.concatenate([incolpos,[outcolpos.max()]])
mi@0:         X = np.concatenate([X, X[:,-1].reshape(X.shape[0],1)], axis=1)
mi@0:     outcolpos = np.concatenate([outcolpos, [outcolpos[-1]]])
mi@0:     # durations (default weights) of input columns)
mi@0:     incoldurs = np.concatenate([np.diff(incolpos), [1]])
mi@0: 
mi@0:     for c in range(noutcols):
mi@0:         firstincol = np.where(incolpos <= outcolpos[c])[0][-1]
mi@0:         firstincolnext = np.where(incolpos < outcolpos[c+1])[0][-1]
mi@0:         lastincol = max(firstincol,firstincolnext)
mi@0:         # default weights
mi@0:         wts = copy.deepcopy(incoldurs[firstincol:lastincol+1])
mi@0:         # now fix up by partial overlap at ends
mi@0:         if len(wts) > 1:
mi@0:             wts[0] = wts[0] - (outcolpos[c] - incolpos[firstincol])
mi@0:             wts[-1] = wts[-1] - (incolpos[lastincol+1] - outcolpos[c+1])
mi@0:         wts = wts * 1. /sum(wts)
mi@0:         Y[:,c] = np.dot(X[:,firstincol:lastincol+1], wts)
mi@0:     # done
mi@0:     return Y
mi@0: 
mi@0: def magnitude(X):
mi@0:     """Magnitude of a complex matrix."""
mi@0:     r = np.real(X)
mi@0:     i = np.imag(X)
mi@0:     return np.sqrt(r * r + i * i);
mi@0: 
mi@0: def json_to_bounds(segments_json):
mi@0:     """Extracts the boundaries from a json file and puts them into 
mi@0:         an np array."""
mi@0:     f = open(segments_json)
mi@0:     segments = json.load(f)["segments"]
mi@0:     bounds = []
mi@0:     for segment in segments:
mi@0:         bounds.append(segment["start"])
mi@0:     bounds.append(bounds[-1] + segments[-1]["duration"]) # Add last boundary
mi@0:     f.close()
mi@0:     return np.asarray(bounds)
mi@0: 
mi@0: def json_bounds_to_bounds(bounds_json):
mi@0:     """Extracts the boundaries from a bounds json file and puts them into 
mi@0:         an np array."""
mi@0:     f = open(bounds_json)
mi@0:     segments = json.load(f)["bounds"]
mi@0:     bounds = []
mi@0:     for segment in segments:
mi@0:         bounds.append(segment["start"])
mi@0:     f.close()
mi@0:     return np.asarray(bounds)
mi@0: 
mi@0: def json_to_labels(segments_json):
mi@0:     """Extracts the labels from a json file and puts them into 
mi@0:         an np array."""
mi@0:     f = open(segments_json)
mi@0:     segments = json.load(f)["segments"]
mi@0:     labels = []
mi@0:     str_labels = []
mi@0:     for segment in segments:
mi@0:         if not segment["label"] in str_labels:
mi@0:             str_labels.append(segment["label"])
mi@0:             labels.append(len(str_labels)-1)
mi@0:         else:
mi@0:             label_idx = np.where(np.asarray(str_labels) == segment["label"])[0][0]
mi@0:             labels.append(label_idx)
mi@0:     f.close()
mi@0:     return np.asarray(labels)
mi@0: 
mi@0: def json_to_beats(beats_json_file):
mi@0:     """Extracts the beats from the beats_json_file and puts them into
mi@0:         an np array."""
mi@0:     f = open(beats_json_file, "r") 
mi@0:     beats_json = json.load(f)
mi@0:     beats = []
mi@0:     for beat in beats_json["beats"]:
mi@0:         beats.append(beat["start"])
mi@0:     f.close()
mi@0:     return np.asarray(beats)
mi@0: 
mi@0: def analyze_results(file):
mi@0:     f = open(file, "r")
mi@0:     lines = f.readlines()
mi@0:     F = []
mi@0:     for line in lines:
mi@0:         F.append(float(line.split("\t")[0]))
mi@0:     f.close()
mi@0:     print np.mean(F)
mi@0: 
mi@0: def compute_ffmc2d(X):
mi@0:     """Computes the 2D-Fourier Magnitude Coefficients."""
mi@0:     # 2d-fft
mi@0:     fft2 = scipy.fftpack.fft2(X)
mi@0: 
mi@0:     # Magnitude
mi@0:     fft2m = magnitude(fft2)
mi@0: 
mi@0:     # FFTshift and flatten
mi@0:     fftshift = scipy.fftpack.fftshift(fft2m).flatten()
mi@0: 
mi@0:     #cmap = plt.cm.get_cmap('hot')
mi@0:     #plt.imshow(np.log1p(scipy.fftpack.fftshift(fft2m)).T, interpolation="nearest", 
mi@0:     #    aspect="auto", cmap=cmap)
mi@0:     #plt.show()
mi@0: 
mi@0:     # Take out redundant components
mi@0:     return fftshift[:fftshift.shape[0]/2+1]