def plotDataset(X, Y, name, img=False, normalizeCols=False,
maxPerClass=np.inf, transforms=None):
print("plotting dataset: " + name + "...")
if normalizeCols:
X = meanNormalizeCols(X)
if img:
imgExamples(X, Y, maxPerClass, transforms)
plt.suptitle(name)
else:
plotExamples(X, Y, maxPerClass, normalize, transforms)
plt.title(name)
suffix = ""
subdir = ""
if img:
suffix += "_img"
subdir += "_img"
if normalizeCols:
suffix += "_colnormalized"
subdir += "_colnormalized"
if maxPerClass < np.inf:
suffix += "_" + str(maxPerClass)
subdir += "_" + str(maxPerClass)
if transforms:
for transform in transforms:
name = '_' + str(transform.__name__)
suffix += name
subdir += name
saveCurrentPlot(name, suffix=suffix, subdir=subdir)
def plotDataset(X,
Y,
name,
img=False,
normalizeCols=False,
maxPerClass=np.inf,
transforms=None):
print("plotting dataset: " + name + "...")
if normalizeCols:
X = meanNormalizeCols(X)
if img:
imgExamples(X, Y, maxPerClass, transforms)
plt.suptitle(name)
else:
plotExamples(X, Y, maxPerClass, normalize, transforms)
plt.title(name)
suffix = ""
subdir = ""
if img:
suffix += "_img"
subdir += "_img"
if normalizeCols:
suffix += "_colnormalized"
subdir += "_colnormalized"
if maxPerClass < np.inf:
suffix += "_" + str(maxPerClass)
subdir += "_" + str(maxPerClass)
if transforms:
for transform in transforms:
name = '_' + str(transform.__name__)
suffix += name
subdir += name
saveCurrentPlot(name, suffix=suffix, subdir=subdir)
def showClusters(name, X, Y, ks, lengths, clusterFactory=makeKMeans, stride=1, normSubSeqs=True):
print("showing clusters for dataset %s" % name)
plt.figure(figsize=(17, 9.5))
plt.rcParams["font.size"] = 18
plotNum = 0
plotRows = len(ks)
plotCols = len(lengths)
n = max(max(lengths), 64) # if we want k > 64, ts must be at least this long
n = min(n, X.shape[1]) # can't be longer than ts
print("Resampling and normalizing...")
X2 = resampleToLengthN(X, n)
X2 = zNormalizeRows(X2)
print("Fitting and plotting...")
for i, k in enumerate(ks):
for j, l in enumerate(lengths):
plotNum += 1
if l > n: # window longer than whole sequence
continue
subseqs = window.sliding_windows_of_rows(X2, l, stride)
if normSubSeqs:
subseqs = zNormalizeRows(subseqs)
clusterer = clusterFactory(subseqs, k)
print subseqs.shape
print subseqs
clusterer.fit(subseqs)
# cluster centers are the actual means
if hasattr(clusterer, 'cluster_centers_'):
centers = clusterer.cluster_centers_
else:
lbls = clusterer.labels_
# grouped = munge.groupXbyY(subseqs, lbls)
grouped = groupXbyY(subseqs, lbls)
centers = map(lambda rows: rows.mean(axis=0), grouped)
plt.subplot(plotRows, plotCols, plotNum)
for center in centers:
plt.plot(np.arange(len(center)), center)
plt.xticks(())
plt.yticks(())
plt.tight_layout()
if i == 0:
plt.title("Window length = %d" % l)
if j == 0:
plt.ylabel("K = %d" % k)
algoName = clusterer.__class__.__name__
if normSubSeqs:
algoName += '-Normalized'
plt.suptitle("Clusters in %s Dataset using %s" % (name, algoName))
plt.tight_layout()
plt.subplots_adjust(left=.03, right=.97, bottom=.01, top=.9, wspace=.02,
hspace=.01)
# all algos for each dataset in one dir, as well as subdir for each algo
saveCurrentPlot("%s_%s.png" % (name, algoName), subdir='cluster')
saveCurrentPlot("%s_%s.png" % (name, algoName),
subdir=os.path.join('cluster', algoName))
plt.close()
def showClusters(name,
X,
Y,
ks,
lengths,
clusterFactory=makeKMeans,
stride=1,
normSubSeqs=True):
print("showing clusters for dataset %s" % name)
plt.figure(figsize=(17, 9.5))
plt.rcParams["font.size"] = 18
plotNum = 0
plotRows = len(ks)
plotCols = len(lengths)
n = max(max(lengths),
64) # if we want k > 64, ts must be at least this long
n = min(n, X.shape[1]) # can't be longer than ts
print("Resampling and normalizing...")
X2 = resampleToLengthN(X, n)
X2 = zNormalizeRows(X2)
print("Fitting and plotting...")
for i, k in enumerate(ks):
for j, l in enumerate(lengths):
plotNum += 1
if l > n: # window longer than whole sequence
continue
subseqs = window.sliding_windows_of_rows(X2, l, stride)
if normSubSeqs:
subseqs = zNormalizeRows(subseqs)
clusterer = clusterFactory(subseqs, k)
print subseqs.shape
print subseqs
clusterer.fit(subseqs)
# cluster centers are the actual means
if hasattr(clusterer, 'cluster_centers_'):
centers = clusterer.cluster_centers_
else:
lbls = clusterer.labels_
# grouped = munge.groupXbyY(subseqs, lbls)
grouped = groupXbyY(subseqs, lbls)
centers = map(lambda rows: rows.mean(axis=0), grouped)
plt.subplot(plotRows, plotCols, plotNum)
for center in centers:
plt.plot(np.arange(len(center)), center)
plt.xticks(())
plt.yticks(())
plt.tight_layout()
if i == 0:
plt.title("Window length = %d" % l)
if j == 0:
plt.ylabel("K = %d" % k)
algoName = clusterer.__class__.__name__
if normSubSeqs:
algoName += '-Normalized'
plt.suptitle("Clusters in %s Dataset using %s" % (name, algoName))
plt.tight_layout()
plt.subplots_adjust(left=.03,
right=.97,
bottom=.01,
top=.9,
wspace=.02,
hspace=.01)
# all algos for each dataset in one dir, as well as subdir for each algo
saveCurrentPlot("%s_%s.png" % (name, algoName), subdir='cluster')
saveCurrentPlot("%s_%s.png" % (name, algoName),
subdir=os.path.join('cluster', algoName))
plt.close()