def main(args):
argp = _argparser().parse_args(args[1:])
### Build the numpy array to be used by the library
if not argp.whitespace_cells:
cell_sep = '\t'
else:
cell_sep = None
# Keep the tokens along to align them later
toks = []
toks_vals = []
for line_num, line in enumerate((l.rstrip('\n') for l in argp.input),
start=1):
try:
if not argp.no_tokens:
tok, vals_str = line.split(cell_sep, 1)
else:
vals_str = line
tok_vals = [float(v) for v in vals_str.split(cell_sep)]
except ValueError:
print >> stderr, ('ERROR: Failed to read input line %s "%s"'
) % (line_num, line, )
print >> stderr, ('ERROR: Is the input perhaps separated by '
'spaces instead of tabs? If so, try the -w flag')
return -1
if not argp.no_tokens:
toks.append(tok)
else:
toks.append(line_num)
toks_vals.append(tok_vals)
toks_vals_array = array(toks_vals)
del toks_vals
### Perform t-SNE (this is heavy stuff)
# TODO: A majority of the below arguments could be program arguments
tsne_output = tsne(toks_vals_array, no_dims=2, perplexity=30, initial_dims=30)
del toks_vals_array
### Re-format and produce TSV output
# Normalise the dimensions onto the range [0, 1]
max_x = max(x for x, _ in tsne_output)
min_x = min(x for x, _ in tsne_output)
max_y = max(y for _, y in tsne_output)
min_y = min(y for _, y in tsne_output)
x_interval = sqrt((max_x - min_x) ** 2)
y_interval = sqrt((max_y - min_y) ** 2)
normalised_tsne_output = (((x - min_x) / x_interval,
(y - min_y) / y_interval) for x, y in tsne_output)
### And write it all out
for token, point in izip(toks, normalised_tsne_output):
argp.output.write('{0}\t{1}\t{2}\n'.format(token, point[0], point[1]))
return 0
def main(argv=None):
if argv is None:
argv = sys.argv[1:]
args = parser.parse_args(argv)
log.info('start parameters: ' + str(args))
log.info('loading data')
titles, x = getData(args.file)
file_size1 = len(titles)
if args.file2 is not None:
titles2, x2 = getData(args.file2)
titles.extend(titles2)
x = np.vstack((x, x2))
# x = scaleData(x)
log.info('performing t-SNE')
out = tsne(x, no_dims=2, perplexity=30, initial_dims=100, use_pca=False)
points = [('green', [
(title, point[0], point[1])
for title, point in zip(titles[:file_size1], out[:file_size1, :])
])]
if args.file2 is not None:
points.append(('gray', [
(title, point[0], point[1])
for title, point in zip(titles[file_size1:], out[file_size1:, :])
]))
# pca = PCA(n_components=2)
# out = pca.fit_transform(x)
# mds = MDS()
# out = mds.fit_transform(x)
log.info('rendering result')
render_points(points, 20, 20)
if args.out:
plt.savefig(args.out, dpi=600)
else:
plt.show()
log.info('finished')
def main(argv=None):
if argv is None:
argv = sys.argv[1:]
args = parser.parse_args(argv)
log.info('start parameters: ' + str(args))
log.info('loading data')
titles, x = getData(args.file)
file_size1 = len(titles)
if args.file2 is not None:
titles2, x2 = getData(args.file2)
titles.extend(titles2)
x = np.vstack((x, x2))
# x = scaleData(x)
log.info('performing t-SNE')
out = tsne(x, no_dims=2, perplexity=30, initial_dims=100, use_pca=False)
points = [('green', [(title, point[0], point[1])
for title, point in zip(titles[:file_size1], out[:file_size1, :])])]
if args.file2 is not None:
points.append(('gray', [(title, point[0], point[1])
for title, point in zip(titles[file_size1:], out[file_size1:, :])]))
# pca = PCA(n_components=2)
# out = pca.fit_transform(x)
# mds = MDS()
# out = mds.fit_transform(x)
log.info('rendering result')
render_points(points, 20, 20)
if args.out:
plt.savefig(args.out, dpi=600)
else:
plt.show()
log.info('finished')
#!/usr/bin/env python
import string, numpy, gzip
o = gzip.open("testdata/english-embeddings.turian.txt.gz", "rb")
titles, x = [], []
for l in o:
toks = string.split(l)
titles.append(toks[0])
x.append([float(f) for f in toks[1:]])
x = numpy.array(x)
#from tsne import tsne
from calc_tsne import tsne
#out = tsne(x, no_dims=2, perplexity=30, initial_dims=30, USE_PCA=False)
#out = tsne(x, no_dims=2, perplexity=30, initial_dims=30, use_pca=False)
out = tsne(x, no_dims=2, perplexity=30, initial_dims=30)
import render
render.render([(title, point[0], point[1]) for title, point in zip(titles, out)], "test-output.rendered.png", width=3000, height=1800)
#!/usr/bin/env python
import string, numpy, gzip
o = gzip.open("testdata/english-embeddings.turian.txt.gz", "rb")
titles, x = [], []
for l in o:
toks = string.split(l)
titles.append(toks[0])
x.append([float(f) for f in toks[1:]])
x = numpy.array(x)
#from tsne import tsne
from calc_tsne import tsne
#out = tsne(x, no_dims=2, perplexity=30, initial_dims=30, USE_PCA=False)
#out = tsne(x, no_dims=2, perplexity=30, initial_dims=30, use_pca=False)
out = tsne(x, no_dims=2, perplexity=30, initial_dims=30)
import render
render.render([(title, point[0], point[1])
for title, point in zip(titles, out)],
"test-output.rendered.png",
width=3000,
height=1800)