def prepare_data(fMRI_file,
subject,
type="glove",
mode="none",
select=True):
brain_activations_1 = genfromtxt(fMRI_file, delimiter=',')
brain_activations = brain_activations_1 - np.mean(brain_activations_1,
axis=0)
brain_activations = np.tanh(brain_activations)
words_1 = []
with open('../data/words', 'r') as f:
reader = csv.reader(f)
words_1 = list(reader)
words = []
words.extend([w[0] for w in words_1])
word_set = list(set(words))
print("number of words: %d " % len(word_set))
selected = np.arange(len(brain_activations_1[0]))
if select == True:
selected_file_name = "general_selected_500_" + subject + ".npy"
if not os.path.isfile(selected_file_name):
selected = select_stable_voxels(brain_activations_1,
word_set,
words,
number_of_trials=6,
size_of_selection=500)
np.save(selected_file_name, selected)
selected = np.load(selected_file_name)
mean_Activations = []
if mode == 'limited':
with open('../data/experimental_wordList.csv', 'r') as f:
word_set = [w[0] for w in list(csv.reader(f))]
words = np.asarray(words)
for word in word_set:
indices = np.where(words == word)[0]
mean_Activations.append(
np.mean(brain_activations[indices, :], axis=0))
words = word_set
if type == 'glove':
wem = WordEmbeddingLayer()
wem.load_filtered_embedding(
"../data/neuro_words_glove_6B_300d") #neuro_words
embedded_words = wem.embed_words(words)
elif type == 'word2vec':
wem = WordEmbeddingLayer()
wem.load_filtered_embedding("../data/neuro_words_word2vec")
embedded_words = wem.embed_words(words)
elif type == 'fasttext':
wem = WordEmbeddingLayer()
wem.load_filtered_embedding("../data/neuro_words_fasttext")
embedded_words = wem.embed_words(words)
elif type == 'lexvec':
wem = WordEmbeddingLayer()
wem.load_filtered_embedding("../data/neuro_words_lexvec")
embedded_words = wem.embed_words(words)
elif type == 'experimental':
embedding_dic, embedded_words = get_word_representation(
type='experimental', words=word_set)
elif type == 'deps':
embedding_dic, embedded_words = get_word_representation(
type='deps', words=word_set)
elif type == 'F25':
embedding_dic, embedded_words = get_word_representation(
type='F25', words=word_set)
elif type == 'non-distributional':
wem = WordEmbeddingLayer()
wem.load_filtered_embedding("../data/neuro_words_cnd")
embedded_words = wem.embed_words(words)
word_representations = np.asarray(embedded_words, np.float32)
return words, np.asarray(word_representations), np.asarray(
mean_Activations)[:, selected]
if not os.path.isfile(selected_file_name):
selected = select_stable_voxels(brain_activations_1,
word_set,
words,
number_of_trials=6,
size_of_selection=500)
np.save(selected_file_name, selected)
selected = np.load(selected_file_name)
mean_Activations = []
words = np.asarray(words)
for word in word_set:
indices = np.where(words == word)[0]
mean_Activations.append(np.mean(brain_activations[indices, :], axis=0))
mean_Activations = np.asarray(mean_Activations)
pdp.compute_and_plot_dists(mean_Activations[:, selected], word_set)
words = word_set
wem = WordEmbeddingLayer()
wem.load_filtered_embedding("../data/neuro_words")
embedded_words = wem.embed_words(word_set)
word_representations = embedded_words
pdp.compute_and_plot_diff_dists(mean_Activations[:, selected],
word_representations, word_set)
lrm.load_model("../glove_all.model")
expSetup_reversed = ExpSetup(learning_rate=0.01,
batch_size=29,
number_of_epochs=2000)
lrm_reversed = tf_LRModel_reversed.LRModel(
y_train.shape[1],
x_train.shape[1],
learning_rate=expSetup_reversed.learning_rate,
hidden_dim=x_train.shape[1],
training_steps=expSetup_reversed.number_of_epochs,
batch_size=expSetup_reversed.batch_size)
lrm_reversed.load_model("../glove_reversed_all_3.model")
wem = WordEmbeddingLayer()
wem.load_embeddings_from_glove_file(
filename="../data/glove.6B/glove.6B.300d.txt", filter=[], dim=300)
#wem.load_filtered_embedding("../data/glove_all_6B_300d") # neuro_words
all_words = [w for w in wem.word2vec.keys()]
all_embedded_words = wem.embed_words(all_words[:len(all_words)])
all_embedded_words = np.asarray(all_embedded_words)
print(all_embedded_words.shape)
word_tree = cKDTree(all_embedded_words)
word_tree_cheat = cKDTree(x_all)
coords = []
with open('../data/coords', 'r') as f:
reader = csv.reader(f)
coords = list(reader)