'Plot the graphs of prediction accuracy of the network vs. fixations',
action="store_true")
args = parser.parse_args()
# Load data
load_data = LoadData()
slide_sequences, slide_selections = load_data.get_RNN_data_selection()
max_sequence_lengths = load_data.get_max_sequence_lengths()
NUM_SUBJECTS = slide_selections.shape[1]
_, _, _, corr_imgs = load_data.get_other_var()
if args.from_scratch:
#Load models
network = Network()
network.load_models("SELECTION")
models = network.get_models(
) #NUM_SLIDES x k model (each slide has k folds)
stratified_kfold = StratifiedKFold(n_splits=k,
shuffle=True,
random_state=seed)
# a database of all probability vs fixation for all slides and all images in each slide
# e.g. all_prob_fix_arr[20][5] will show the whole progression of probability vs fixation for slide 21, image 6
all_prob_fix_arr = []
# a database of all prediction accuracy vs fixation for all slides
# e.g. all_acc_arr[20] returns an array of max_sequence_lengths[20] elements
all_acc_fix_arr = []
model_type_name = "CORRECT"
print("model_type_name: ", model_type_name)
load_data = LoadData()
if model_type_name == "SELECTION":
slide_sequences, slide_labels = load_data.get_RNN_data_selection()
else:
slide_sequences, slide_labels = load_data.get_RNN_data()
max_sequence_lengths = load_data.get_max_sequence_lengths()
args = parser.parse_args()
network = Network()
network.load_models(model_type_name)
model = network.get_models()
method = int(args.method)
if method == 1:
ablated_sequences = modify_all_sequences(slide_sequences, 0, 'shuffle', 'post')
elif method == 2:
ablated_sequences = modify_all_sequences(slide_sequences, 'half', 'discard', 'post')
elif method == 3:
ablated_sequences = modify_all_sequences(slide_sequences, 'half', 'discard', 'pre')
elif method == 4:
ablated_sequences = modify_all_sequences(slide_sequences, 'half', 'shuffle', 'pre')
elif method == 5:
ablated_sequences = modify_all_sequences(slide_sequences, 'half', 'shuffle', 'post')
elif method == 6:
ablated_sequences = modify_all_sequences(slide_sequences, 'half', 'shuffle', 'post') #shuffle second half