def main(outfile=''):
manifolds = ['euclidean',
# 'transe',
'poincare',
'lorentz']
dimensions = [5, 10, 20, 50, 100, 200]
js = {}
for mani in manifolds:
for dim in dimensions:
key_json = '%s%d' % (mani, dim)
txtdir = './emb150txt/%s%d/' % (mani, dim)
txtfile = txtdir + ('%s%d.txt' % (mani, dim))
print('loading %s...' % txtfile)
keyvalues = PoincareKeyedVectors.load_word2vec_format(txtfile,
fvocab=None,
binary=False,
encoding='utf8',
unicode_errors='strict',
limit=None,
datatype=np.float64)
pr = generate_data()
actual_dim = dim
if mani == 'lorentz':
actual_dim += 1
eva = ReconstructionEvaluation(pr, keyvalues)
# print('filename=%s: ' % txtfile)
res = eva.evaluate()
js[key_json] = res
with open(outfile, 'w') as fp:
json.dump(js,fp)
def __init__(self, ns):
"""
:param ns: ['molecular_function', 'biological_process', 'cellular_component']
"""
self._aspect = aspect = GoAspect(ns)
self._graph = G = get_ontology_graph(ns)
classes = list(reversed(list(nx.topological_sort(G))))
self.root = root = classes[0]
self._levels = self.bfs(root)
self._levels = levels = dict()
for node, lvl in nx.shortest_path_length(G, target=root).items():
if lvl in levels:
levels[lvl].append(node)
else:
levels[lvl] = [node]
self._mlb = MultiLabelBinarizer().fit([classes])
key_val = [(go, i) for i, go in enumerate(classes)]
self.go2ix = {k: v for k, v in key_val}
self.ix2go = {v: k for k, v in key_val}
emb_fname = os.path.join('%s/%s-poincare-dim%d-epochs%d.emb' %
(DATA_ROOT, aspect, dim, num_epochs))
if os.path.exists(emb_fname):
self._kv = PoincareKeyedVectors.load(emb_fname)
else:
self._kv = embedding(ns, emb_fname)
def __init__(self, params):
super().__init__(params)
self.wiktionary = self.__get_wiktionary(params['wiki_path'])
self.wiki_model = KeyedVectors.load_word2vec_format(params['wiki_vectors_path'], binary=False)
self.node2vec = KeyedVectors.load_word2vec_format(params["node2vec_path"], binary=False)
self.n = params['n']
self.projection = ProjectionVectorizer(self.w2v_data, params["projection_path"])
self.poincare_model = PoincareKeyedVectors.load_word2vec_format(params["poincare_path"], binary=False)
self.n = params["n"]
self.delete_bracets = re.compile(r"\(.+?\)")
if params['language'] == 'ru':
self.pattern = re.compile("[^А-я \-]")
else:
self.pattern = re.compile("[^A-z \-]")
def load_poincare_model(path, word2vec_format=True, binary=False):
""" Load a Poincare embedding model.
:param path: path of the file of the pre-trained Poincare embedding model
:param word2vec_format: whether to load from word2vec format (default: True)
:param binary: binary format (default: False)
:return: a pre-trained Poincare embedding model
:type path: str
:type word2vec_format: bool
:type binary: bool
:rtype: gensim.models.poincare.PoincareKeyedVectors
"""
if word2vec_format:
return PoincareKeyedVectors.load_word2vec_format(path, binary=binary)
else:
return PoincareModel.load(path).kv
def main(poincare=''):
from gensim.models.poincare import PoincareModel
pm = PoincareModel([], size=300, dtype=np.float64)
emb = PoincareKeyedVectors.load_word2vec_format(poincare,
fvocab=None,
binary=False,
encoding='utf8',
unicode_errors='strict',
limit=None,
datatype=np.float64)
pm.kv = emb
pm.save('w2v_poincare.pickle', pickle_protocol=4)
pm2 = PoincareModel.load('w2v_poincare.pickle')
pm2.train(10000, batch_size=10, print_every=1, check_gradients_every=None)
pm2.save('w2v_poincare_after_train.pickle', pickle_protocol=4)
def poincare_distance_heatmap(origin_point, x_range=(-1.0, 1.0), y_range=(-1.0, 1.0), num_points=100):
"""Create a heatmap of Poincare distances from `origin_point` for each point (x, y),
where x and y lie in `x_range` and `y_range` respectively, with `num_points` points chosen uniformly in both ranges.
Parameters
----------
origin_point : tuple (int, int)
(x, y) from which distances are to be measured and plotted.
x_range : tuple (int, int)
Range for x-axis from which to choose `num_points` points.
y_range : tuple (int, int)
Range for y-axis from which to choose `num_points` points.
num_points : int
Number of points to choose from `x_range` and `y_range`.
Notes
-----
Points outside the unit circle are ignored, since the Poincare distance is defined
only for points inside the circle boundaries (exclusive of the boundary).
Returns
-------
:class:`plotly.graph_objs.Figure`
Plotly figure that contains plot
"""
epsilon = 1e-8 # Can't choose (-1.0, -1.0) or (1.0, 1.0), distance undefined
x_range, y_range = list(x_range), list(y_range)
if x_range[0] == -1.0 and y_range[0] == -1.0:
x_range[0] += epsilon
y_range[0] += epsilon
if x_range[0] == 1.0 and y_range[0] == 1.0:
x_range[0] -= epsilon
y_range[0] -= epsilon
x_axis_values = np.linspace(x_range[0], x_range[1], num=num_points)
y_axis_values = np.linspace(x_range[0], x_range[1], num=num_points)
x, y = np.meshgrid(x_axis_values, y_axis_values)
all_points = np.dstack((x, y)).swapaxes(1, 2).swapaxes(0, 1).reshape(2, num_points ** 2).T
norms = np.linalg.norm(all_points, axis=1)
all_points = all_points[norms < 1]
origin_point = np.array(origin_point)
all_distances = PoincareKeyedVectors.poincare_dists(origin_point, all_points)
distances = go.Scatter(
x=all_points[:, 0],
y=all_points[:, 1],
mode='markers',
marker=dict(
size='9',
color=all_distances,
colorscale='Viridis',
showscale=True,
colorbar=go.ColorBar(
title='Poincare Distance'
),
),
text=[
'Distance from (%.2f, %.2f): %.2f' % (origin_point[0], origin_point[1], d)
for d in all_distances],
name='', # To avoid the default 'trace 0'
)
origin = go.Scatter(
x=[origin_point[0]],
y=[origin_point[1]],
name='Distance from (%.2f, %.2f)' % (origin_point[0], origin_point[1]),
mode='markers+text',
marker=dict(
size='10',
color='rgb(200, 50, 50)'
)
)
layout = go.Layout(
width=900,
height=800,
showlegend=False,
title='Poincare Distances from (%.2f, %.2f)' % (origin_point[0], origin_point[1]),
hovermode='closest',
)
return go.Figure(data=[distances, origin], layout=layout)
def setUp(self):
self.vectors = PoincareKeyedVectors.load_word2vec_format(
datapath('poincare_vectors.bin'), binary=True)
def setUp(self):
self.vectors = PoincareKeyedVectors.load_word2vec_format(datapath('poincare_vectors.bin'), binary=True)
def poincare_distance_heatmap(origin_point, x_range=(-1.0, 1.0), y_range=(-1.0, 1.0), num_points=100):
"""Create a heatmap of Poincare distances from `origin_point` for each point (x, y),
where x and y lie in `x_range` and `y_range` respectively, with `num_points` points chosen uniformly in both ranges.
Parameters
----------
origin_point : tuple (int, int)
(x, y) from which distances are to be measured and plotted.
x_range : tuple (int, int)
Range for x-axis from which to choose `num_points` points.
y_range : tuple (int, int)
Range for y-axis from which to choose `num_points` points.
num_points : int
Number of points to choose from `x_range` and `y_range`.
Notes
-----
Points outside the unit circle are ignored, since the Poincare distance is defined
only for points inside the circle boundaries (exclusive of the boundary).
Returns
-------
:class:`plotly.graph_objs.Figure`
Plotly figure that contains plot
"""
epsilon = 1e-8 # Can't choose (-1.0, -1.0) or (1.0, 1.0), distance undefined
x_range, y_range = list(x_range), list(y_range)
if x_range[0] == -1.0 and y_range[0] == -1.0:
x_range[0] += epsilon
y_range[0] += epsilon
if x_range[0] == 1.0 and y_range[0] == 1.0:
x_range[0] -= epsilon
y_range[0] -= epsilon
x_axis_values = np.linspace(x_range[0], x_range[1], num=num_points)
y_axis_values = np.linspace(x_range[0], x_range[1], num=num_points)
x, y = np.meshgrid(x_axis_values, y_axis_values)
all_points = np.dstack((x, y)).swapaxes(1, 2).swapaxes(0, 1).reshape(2, num_points ** 2).T
norms = np.linalg.norm(all_points, axis=1)
all_points = all_points[norms < 1]
origin_point = np.array(origin_point)
all_distances = PoincareKeyedVectors.poincare_dists(origin_point, all_points)
distances = go.Scatter(
x=all_points[:, 0],
y=all_points[:, 1],
mode='markers',
marker=dict(
size='9',
color=all_distances,
colorscale='Viridis',
showscale=True,
colorbar=go.ColorBar(
title='Poincare Distance'
),
),
text=[
'Distance from (%.2f, %.2f): %.2f' % (origin_point[0], origin_point[1], d)
for d in all_distances],
name='', # To avoid the default 'trace 0'
)
origin = go.Scatter(
x=[origin_point[0]],
y=[origin_point[1]],
name='Distance from (%.2f, %.2f)' % (origin_point[0], origin_point[1]),
mode='markers+text',
marker=dict(
size='10',
color='rgb(200, 50, 50)'
)
)
layout = go.Layout(
width=900,
height=800,
showlegend=False,
title='Poincare Distances from (%.2f, %.2f)' % (origin_point[0], origin_point[1]),
hovermode='closest',
)
return go.Figure(data=[distances, origin], layout=layout)
def __init__(self, params):
super().__init__(params)
self.poincare_model = PoincareKeyedVectors.load_word2vec_format(params["poincare_path"], binary=False)
self.n = params["n"]
from gensim.models.poincare import PoincareModel, PoincareKeyedVectors, PoincareRelations
from gensim.viz.poincare import poincare_distance_heatmap
from tensorflow.keras.layers import Embedding
wordnet_mamal_file_path = '/Users/pankaj/dev/git/smu/nlp337/data/mamals.tsv'
relations = PoincareRelations(wordnet_mamal_file_path, delimiter='\t')
model = PoincareModel(train_data=relations, size=2, burn_in=0)
model.train(epochs=2, print_every=500)
pcv = PoincareKeyedVectors(vector_size=20)
poincare_distance_heatmap((0, 0),
x_range=(-1.0, 1.0),
y_range=(-1.0, 1.0),
num_points=100)
