import argparse
import logging
import numpy as np
from sklearn import linear_model
from attalos.dataset.dataset import Dataset
from attalos.dataset.transformers.onehot import OneHot
from attalos.evaluation.evaluation import Evaluation
import attalos.util.log.log as l
logger = l.getLogger(__name__)
def get_xy(dataset, tag_transformer=None):
x = []
y = []
for idx in dataset:
image_feats, text_feats = dataset.get_index(idx)
if tag_transformer:
text_feats = tag_transformer.get_multiple(text_feats)
x.append(image_feats)
y.append(text_feats)
return np.asarray(x), np.asarray(y)
def train(train_dataset, test_dataset=None, tag_transformer=None, n_jobs=-1):
x, targets = get_xy(train_dataset, tag_transformer=tag_transformer)
model = linear_model.LinearRegression(n_jobs=n_jobs)
logger.info("Training model.")
model.fit(x, targets)
import numpy as np
import tensorflow as tf
from attalos.imgtxt_algorithms.approaches.base import AttalosModel
from attalos.util.transformers.onehot import OneHot
from attalos.imgtxt_algorithms.correlation.correlation import construct_W
from attalos.imgtxt_algorithms.util.negsamp import NegativeSampler
import attalos.util.log.log as l
logger = l.getLogger(__name__)
class FastZeroTagModel(AttalosModel):
"""
Create a tensorflow graph that finds the principal direction of the target word embeddings
(with negative sampling), using the loss function from "Fast Zero-Shot Image Tagging".
"""
def __init__(self, wv_model, datasets, **kwargs):
self.wv_model = wv_model
self.one_hot = OneHot(datasets, valid_vocab=wv_model.vocab)
word_counts = NegativeSampler.get_wordcount_from_datasets(datasets, self.one_hot)
self.negsampler = NegativeSampler(word_counts)
self.w = construct_W(wv_model, self.one_hot.get_key_ordering()).T
self.learning_rate = kwargs.get("learning_rate", 0.0001)
self.optim_words = kwargs.get("optim_words", True)
self.hidden_units = kwargs.get("hidden_units", "200")
self.use_batch_norm = kwargs.get("use_batch_norm",False)
self.opt_type = kwargs.get("opt_type","adam")
if self.hidden_units=='0':
self.hidden_units=[]
else:
self.hidden_units = [int(x) for x in self.hidden_units.split(',')]