res_size = 0
if bidirectional == "True":
total_res_size = res_size * 2
else:
total_res_size = res_size
a = float(args.leak_rate) # leaking rate
res_sparsity = float(args.res_sparsity)
only_open_class = args.only_open_class
training_iterations = int(args.training_iterations)
save_path = args.save_path
embeddings = load_embeddings(embeddings_model) # load the embeddings
f_sensekey2synset = cPickle.load(open(
sensekey2synset, "rb")) # get the mapping between synset keys and IDs
train_data, known_lemmas = read_data(
train_data_path, f_sensekey2synset,
only_open_class) # read the training data
lemma2synset = get_lemma2syn(dictionary)
lemma2id, synset2id = get_lemma_synset_maps(lemma2synset, known_lemmas)
if softmax == "True":
output_size = len(synset2id)
else:
output_size = embeddings_size
input_size = embeddings_size * window_size
# generate the ESN reservoir
name_add = str(res_size) + '_' + str(res_sparsity) + '_' + str(a)
random.seed(42)
Wout = (random.rand(output_size, total_res_size + input_size) - 0.5) * 1.0
if softmax == "True":
import os import matplotlib import matplotlib.pyplot as plt import numpy as np import cv2 import datetime from sklearn.model_selection import train_test_split from sklearn import preprocessing import pdb from preprocess_data import read_data ''' !!!the KTH dataset may have 599 files,I add one duplicate video in the folder having 99 images ''' X_tr = read_data() X_train = np.array(X_tr) #convert the frames read into array num_samples = len(X_train) #600 label = np.ones((num_samples, ), dtype=int) #each class has 100 videos label[0:100] = 0 label[100:200] = 1 label[200:300] = 2 label[300:400] = 3 label[400:500] = 4 label[500:] = 5 y_train = label #600 num_samples img_rows, img_cols, img_depth = 32, 32, 15
bidirectional = args.bidirectional
if use_reservoirs == "True":
res_size = int(args.res_size)
else:
res_size = 0
if bidirectional == "True":
total_res_size = res_size * 2
else:
total_res_size = res_size
only_open_class = args.only_open_class
save_path = args.save_path
embeddings = load_embeddings(embeddings_model) # load the embeddings
f_sensekey2synset = cPickle.load(open(
sensekey2synset, "rb")) # get the mapping between synset keys and IDs
test_data = read_data(test_data_path, f_sensekey2synset, only_open_class)
inSize = embeddings_size * window_size
outSize = embeddings_size
with open(save_path, "rb") as input_file:
_ = cPickle.load(input_file)
resSparsity = cPickle.load(input_file)
a = cPickle.load(input_file)
Wout = cPickle.load(input_file)
if use_reservoirs == "True":
Win_fw = cPickle.load(input_file)
Win_bw = cPickle.load(input_file)
W_fw = cPickle.load(input_file)
W_bw = cPickle.load(input_file)
G_fw = cPickle.load(input_file)
if bidirectional == "True":
total_res_size = res_size * 2
else:
total_res_size = res_size
only_open_class = args.only_open_class
save_path = args.save_path
f_syn2gloss = args.syn2gloss
if f_syn2gloss != "None":
syn2gloss = pickle.load(open(f_syn2gloss, "rb"))
else:
syn2gloss = None
error_log = args.error_log
embeddings = load_embeddings(embeddings_model) # load the embeddings
f_sensekey2synset = cPickle.load(open(sensekey2synset, "rb")) # get the mapping between synset keys and IDs
test_data, known_lemmas = read_data(test_data_path, f_sensekey2synset, only_open_class)
input_size = embeddings_size * window_size
output_size = embeddings_size
pickled_files = ""
for f in os.listdir(save_path):
if f.endswith(".cpickle"):
pickled_files = f
synset2id = None
with open(os.path.join(save_path, pickled_files), "rb") as input_file:
_ = cPickle.load(input_file)
resSparsity = cPickle.load(input_file)
a = cPickle.load(input_file)
if softmax == "True":