def __init__(self,
train_data=None,
test_data=None,
save_path="doc2vec.pt",
log=utils.get_logging()):
self.model = gensim.models.doc2vec.Doc2Vec(size=50,
min_count=2,
iter=55)
self.train_data = train_data
self.test_data = test_data
self.save_path = save_path
self.log = log
#!/usr/bin/env python
# encoding: utf-8
from __future__ import print_function
import argparse
from functools import partial
import numpy as np
import multiprocessing as mp
import torch
import figet
from figet import utils
log = utils.get_logging()
def tune(baseline, dist, type_, num_types, init_threshold):
idx2threshold = {idx: init_threshold for idx in xrange(num_types)}
pool = mp.Pool(processes=8)
func = partial(search_threshold,
init_threshold=init_threshold,
num_types=num_types,
dist=dist,
type_=type_,
baseline=baseline)
for idx, best_t in pool.map(func, xrange(num_types)):
idx2threshold[idx] = best_t
return idx2threshold
def search_threshold(idx, init_threshold, num_types, dist, type_, baseline):
from pyflann import *
from pyflann.exceptions import FLANNException
import numpy as np
from figet.utils import get_logging
from figet.Constants import COARSE_FLAG, FINE_FLAG, UF_FLAG
from figet.hyperbolic import poincare_distance
import torch
from operator import itemgetter
log = get_logging()
cos_sim_func = torch.nn.CosineSimilarity(dim=0)
def cosine_distance(a, b):
return 1 - cos_sim_func(a, b)
class kNN(object):
def __init__(self, type2vec, type_vocab, metric):
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
self.type2vec = type2vec.to(self.device).type(torch.float)
self.type_vocab = type_vocab
self.hyperbolic = metric == "hyperbolic"
self.query_factor = 25 if self.hyperbolic else 1
self.neighs_per_granularity = {
COARSE_FLAG: 1,
FINE_FLAG: 1,
UF_FLAG: 3
}