validation_steps=64,
callbacks=[tb_callback, cp_callback])
print('TRAINING COMPLETE')
model.save(MODEL_STRUCT_PATH)
for fp in glob(os.path.join(TEST_DATA_PATH, '*', '7000.png')):
print(fp)
(fn, _) = os.path.splitext(fp)
arr = numpy.array(
load_img(fp,
target_size=(CHAR_IMG_HEIGHT, CHAR_IMG_WIDTH),
grayscale=False,
color_mode='rgb',
interpolation='nearest'))
a = get_activations(model, [[arr]], auto_compile=True)
rp = os.path.join(KERACT_PATH, relative_path(fn, TEST_DATA_PATH))
display_activations(a, directory=rp, save=True)
print(f'VISUALIZATION SAVED: {rp}')
print('DONE')
yp = model.predict_generator(test_generator)
yp = numpy.argmax(yp, axis=1)
print('CONFUSION MATRIX:')
print(confusion_matrix(test_generator.classes, yp))
print('Classification Report')
print(
classification_report(test_generator.classes,
yp,
def test_nodes_and_layer_name(self):
model, x = dummy_model_and_inputs()
self.assertRaises(
ValueError, lambda: get_activations(
model, x, nodes_to_evaluate=[], layer_names='unknown'))
def fun_to_test(auto_compile):
get_activations(model, x, auto_compile=auto_compile)
lb = LabelEncoder()
pred = np.round(model.predict(x_test, verbose=1))
pred = pred.squeeze().argmax(axis=1)
new_y_test = y_test.astype(int)
mtx = confusion_matrix(new_y_test, pred)
labels = [
'Angry', 'Disgusted', 'Fearful', 'Happy', 'Neutral', 'Sad', 'Surprised'
]
# labels = ['Guilt', 'Disgust', 'Happy', 'Fear', 'Anger', 'Surprise', 'Sad']
h = plt.figure()
sb.heatmap(mtx,
annot=True,
fmt='d',
yticklabels=labels,
xticklabels=labels,
cbar=False)
plt.title('Confusion matrix')
h.savefig("Network/Confusion.pdf", bbox_inches='tight')
# %%
x = np.expand_dims(x_test[2, :, :], 0)
x.shape
l_weights = keract.get_activations(model, x, layer_names='activation')
plt.figure()
plt.plot(np.squeeze(l_weights['activation']))
# %%
from keras.layers import Dense
from keras.layers.recurrent import LSTM
from keras.models import Sequential
import keract
import utils
from data import get_mnist_data, num_classes
if __name__ == '__main__':
x_train, y_train, _, _ = get_mnist_data()
# (60000, 28, 28, 1) to ((60000, 28, 28)
# LSTM has (batch, time_steps, input_dim)
x_train = x_train.squeeze()
model = Sequential()
model.add(LSTM(16, input_shape=(28, 28)))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
metrics=['accuracy'])
utils.print_names_and_shapes(keract.get_activations(model, x_train[:128]))
utils.print_names_and_shapes(
keract.get_gradients_of_trainable_weights(model, x_train[:128],
y_train[:128]))
utils.print_names_and_shapes(
keract.get_gradients_of_activations(model, x_train[:128],
y_train[:128]))
def extract_features_to_pd():
# 1. Load Data and CNN Model
DTL = np.load(
os.path.join(cf.DROPBOX_DIRECTORY, 'Data', 'OPM', 'FinalData',
'Individual Datasets', 'bisp_dtl.npy'))
bisp_df = pd.read_pickle(
os.path.join(cf.DROPBOX_DIRECTORY, 'Data', 'OPM', 'FinalData',
'Individual Datasets', 'bisp_dtl_uids.pkl'))
model = load_model(cf.CNN_FILENAME)
# 2. Extract features
layer_name = 'fc1'
#DTL.shape
#DTL[0].shape
i = 0
DTL_i = DTL[(i):(i + 1)]
l1 = DTL_i[0, :, :, 0]
show(l1)
activations = get_activations(model, DTL[0:1])
import keract
keract.display_activations(activations,
cmap=None,
save=False,
directory='.',
data_format='channels_last',
fig_size=(24, 24),
reshape_1d_layers=False)
keract.display_heatmaps(activations, DTL[0:1], save=False)
a.shape
DTL_p = preprocess_input(DTL) # Preprocess image data
#DTL_p = DTL_p[1:5,:,:,:] # for testing
# Generate feature extractor using trained CNN
feature_extractor = Model(
inputs=model.inputs,
outputs=model.get_layer(name=layer_name).output,
)
features = feature_extractor.predict(DTL_p)
# 3. Create and format pandas DataFrame
df = pd.DataFrame(features).add_prefix('cnn_feat_')
df['uid'] = bisp_df.uid
# 4. Export
df.to_pickle(
os.path.join(cf.DROPBOX_DIRECTORY, 'Data', 'OPM', 'FinalData',
'Individual Datasets', 'bisp_cnn_features_all.pkl'))
df.to_csv(
os.path.join(cf.DROPBOX_DIRECTORY, 'Data', 'OPM', 'FinalData',
'Individual Datasets', 'bisp_cnn_features_all.csv'))
eight_indices = np.where(trainlabels == 8)[0]
nine_indices = np.where(trainlabels == 9)[0]
indices = [
zero_indices, one_indices, two_indices, three_indices, four_indices,
five_indices, six_indices, seven_indices, eight_indices, nine_indices
]
#keract_target=target_test[:1]
meanarraylist = []
for anum in indices:
sumarray = np.zeros(128)
for anindex in anum:
keract_inputs = input_train[anindex]
activations = get_activations(model,
keract_inputs.reshape(1, 28, 28, 1),
layer_names='dense128',
output_format='simple')
sumarray = np.add(sumarray, activations['dense128'])
meanarray = sumarray / len(anum)
meanarraylist.append(meanarray)
#print(meanarraydict)
#activations=get_activations(model, keract_inputs)
#activations=get_activations(model, keract_inputs, layer_names='dense128', output_format='simple')
#display_activations(activations, cmap='gray', save=False)
#heatmaps
from keract import display_heatmaps
#display_heatmaps(activations, keract_inputs, save=False)
labels = ['Guilt', 'Disgust', 'Happy', 'Neutral', 'Anger', 'Surprise', 'Sad']
history_pred = []
hist_time = []
while True:
data = np.frombuffer(stream.read(CHUNK), dtype=np.float32)
x_infer = input_prep(data, smile)
pred = model.predict(x_infer)
predi = pred.argmax(axis=1)
history_pred = np.append(history_pred, predi[0])
# hist_time = np.append(hist_time, dtime.now().strftime('%H:%M:%S'))
print(labels[predi[0]] + " -- (raw data peak: " + str(max(data)) + ")")
# GET ACTIVATIONS
layername = 'activation_3'
l_weights = keract.get_activations(model, x_infer, layer_names=layername)
w_values = np.squeeze(l_weights[layername])
# SEND TO MQTT BrOKER
client.publish('hiper/labinter99_ita', labels[predi[0]])
for k in range(len(labels)):
topic_pub = "hiper/labinter_ita_" + labels[k]
# client.subscribe(topic_pub)
client.publish(topic_pub, str(w_values[k]))
# SEND TO MQTT BrOKER
# for k in range(len(labels)):
# mqtt_client.publish_single(float(w_values[k]), topic=labels[k])
# plot
# clear_output(wait=True)
import numpy as np
from Udacity import utils
from matplotlib import pyplot as plt
save_path = 'feature_maps'
img_path = r'D:\Eitan_Netanel\Records\Normal_3\IMG\center_2018_11_30_16_43_27_723.jpg'
model_path = r'C:\Users\netanelgip\PycharmProjects\CarNet\Udacity\normal_turns_reducelr_elu_lr0.6_batch20_saveall\model-016.h5'
# Load the image. the network expects RGB images
image = cv2.imread(img_path)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = utils.preprocess(image) # apply the preprocessing
model = load_model(model_path)
model.summary()
activations = get_activations(model, [image])
layer_names = list(activations.keys())
activation_maps = list(activations.values())
batch_size = activation_maps[0].shape[0]
assert batch_size == 1, 'One image at a time to visualize.'
for i, activation_map in enumerate(activation_maps):
print('Displaying activation map {}'.format(i))
shape = activation_map.shape
layer_dir = path.join(save_path,
layer_names[i]).replace('/', '').replace(':', '')
if not path.exists(layer_dir):
os.mkdir(layer_dir)
if len(shape) == 4:
# image = origin_image.crop((0, 0, 224, 224))
img = image.load_img(img_path, target_size=(image_size, image_size))
img = img_to_array(img)
img = img.reshape((1, img.shape[0], img.shape[1], img.shape[2]))
img = preprocess_input(img)
yhat = model.predict(img)
label = decode_predictions(yhat)
label = label[0][0]
print('{} ({})'.format(label[1], label[2] * 100))
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
activations = keract.get_activations(model, img, layer_name='block5_conv3')
first = activations.get('block5_conv3')
# keract.display_activations(activations)
# keract.display_heatmaps(activations, input_image=image)
grad_trainable_weights = keract.get_gradients_of_activations(model, img, yhat, layer_name='block5_conv3')
print(grad_trainable_weights['block5_conv3'].shape)
grad_trainable_weights = tf.convert_to_tensor(grad_trainable_weights['block5_conv3'])
pooled_grads = K.mean(grad_trainable_weights, axis=(0, 1, 2))
# 我们计算相类输出值关于特征图的梯度。然后，我们沿着除了通道维度之外的轴对梯度进行池化操作。最后，我们用计算出的梯度值对输出特征图加权。
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
print(model.summary())
x_train, y_train, x_test, y_test = MNIST.get_mnist_data()
# checking that the accuracy is the same as before 99% at the first epoch.
# test_loss, test_acc = model.evaluate(x_test, y_test, verbose=0, batch_size=128)
# print('')
# assert test_acc > 0.98
utils.print_names_and_shapes(
keract.get_activations(model, x_test[0:200])) # with 200 samples.
utils.print_names_and_shapes(
keract.get_gradients_of_trainable_weights(model, x_train[0:10],
y_train[0:10]))
utils.print_names_and_shapes(
keract.get_gradients_of_activations(model, x_train[0:10],
y_train[0:10]))
a = keract.get_activations(model, x_test[0:1]) # with just one sample.
keract.display_activations(a, directory='mnist_activations', save=True)
# import numpy as np
# import matplotlib.pyplot as plt
# plt.imshow(np.squeeze(x_test[0:1]), interpolation='None', cmap='gray')
else:
x_train, y_train, x_test, y_test = MNIST.get_mnist_data()
# Fit data
autoencoder.fit(input_train,
input_train,
epochs=no_epochs,
batch_size=batch_size,
validation_split=validation_split)
# =============================================
# Take a sample for visualization purposes
# =============================================
input_sample = input_test[:1]
reconstruction = autoencoder.predict([input_sample])
# =============================================
# Visualize input-->reconstruction
# =============================================
fig, axes = plt.subplots(1, 2)
fig.set_size_inches(6, 3.5)
input_sample_reshaped = input_sample.reshape((img_width, img_height))
reconsstruction_reshaped = reconstruction.reshape((img_width, img_height))
axes[0].imshow(input_sample_reshaped)
axes[0].set_title('Original image')
axes[1].imshow(reconsstruction_reshaped)
axes[1].set_title('Reconstruction')
plt.show()
# =============================================
# Visualize encoded state with Keract
# =============================================
activations = get_activations(autoencoder, input_sample)
display_activations(activations, cmap="gray", save=False)
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(10, kernel_size=(5, 5), activation='relu'))
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dense(no_classes, activation='softmax'))
# Compile the model
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
# Fit data to model
model.fit(input_train, target_train,
batch_size=batch_size,
epochs=no_epochs,
verbose=verbosity,
validation_split=validation_split)
# Generate generalization metrics
score = model.evaluate(input_test, target_test, verbose=0)
print(f'Test loss: {score[0]} / Test accuracy: {score[1]}')
# =============================================
# Keract visualizations
# =============================================
from keract import get_activations, display_activations, display_heatmaps
keract_inputs = input_test[:1]
keract_targets = target_test[:1]
activations = get_activations(model, keract_inputs)
display_activations(activations, cmap="gray", save=False)
display_heatmaps(activations, keract_inputs, save=False)
#loads the model from the saved model file
json_file = open('model.json', 'r')
mapping = {'LeakyRelu': LeakyRelu}
loaded_model_json = json_file.read()
json_file.close()
model = model_from_json(loaded_model_json, mapping)
# load weights into new model
model.load_weights("model.h5")
#keract requires model compilation
model.compile(loss="mse", optimizer="adam")
while True:
#generates the noise to be fed into the model
noise = np.random.normal(0, 1, (1, 100))
#shows the reshape layer as it has image output
activations = get_activations(model, noise, model.layers[4].name)
display_activations(activations, cmap="gray")
for layer in model.layers:
#shows only the batch norm layers to avoid seeing conv then batch norm when they are quite similar
if "norm" in layer.name:
activations = get_activations(model, noise, layer.name)
display_activations(activations, cmap="gray")
#the last layer doesn't have any batch norm but we want to see it anyway
output = reverse_tanh(model.predict(noise)[0])
plt.imshow(output)
plt.show()
#%%
from keras.models import load_model
from keract import get_activations, display_activations, display_heatmaps
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
#%%
path = Path("/home/killaarsl/Documents/racebaandemo/ADR")
model = load_model(str(path / "tempmodel.h5"))
image = plt.imread(str(path / "434_4_2.png"))
image = image[:, :, :3]
image = np.expand_dims(image, axis=0)
activations = get_activations(model, image)
#%%
display_activations(activations, cmap="gray", save=True)
#%%
display_heatmaps(activations, image, save=False)
patience=15,
verbose=1,
restore_best_weights=True)
# model.fit(
# x=x_train,
# y=y_train,
# verbose=1,
# batch_size=batch_size,
# epochs=epochs,
# callbacks=[plateau_callback, es_callback],
# validation_data=(x_test, y_test))
# model.save_weights(filepath=data_model_path)
model.load_weights(filepath=data_model_path)
# score = model.evaluate(
# x_test,
# y_test,
# verbose=0,
# batch_size=batch_size)
# print("Test performance: ", score)
grads = keract.get_gradients_of_activations(model,
x_test[[12]],
y_test[[12]],
layer_name='heatmap1')
keract.display_heatmaps(grads, x_test[12] * 255.0)
activations = keract.get_activations(model, x_test[[12]])
keract.display_activations(activations)
cmap="gray")
# create an axes on the right side of ax. The width of cax will be 5%
# of ax and the padding between cax and ax will be fixed at 0.05 inch.
divider = make_axes_locatable(ax)
cax = divider.append_axes("bottom", size="5%", pad=0.05)
plt.colorbar(im, cax=cax, orientation="horizontal")
plt.show()
import keract
# Getting all the activations using keract
activations = keract.get_activations(model,
data_events,
layer_names=["Activation"])
#activations = keract.get_activations(model, data_events, layer_names=["activation_133"])
# Plotting using keract
keract.display_heatmaps(
activations, data_events, cmap="gray", save=True,
directory='picss') # to save the files use: save=True, directory='pics' f
#activations = keract.get_activations(model, data_events, layer_names=["conv2d_399", "conv2d_400", "conv2d_401", "max_pooling2d_133"])
activations = keract.get_activations(model,
data_events,
layer_names=[
"TimeConv0", "Dropout0", "TimeConv1",
"Dropout1", "StationConv", "Dropout2",
"Pooling", "DenseFlipout"
])
# print("shape is",acts.shape)
# print(layer_name, acts.shape, end=' ')
#
# if acts.shape[0] != 1:
# print('-> Skipped. First dimension is not 1.')
# continue
# if len(acts.shape) <= 2:
# print('-> Skipped. 2D Activations.')
# continue
report = metrics.classification_report(y_true=y_preds_thresh, y_pred=np.asarray(lbls), output_dict =True)
print(report)
matrix = metrics.confusion_matrix(y_preds_thresh, np.asarray(lbls))
#print(matrix)
act= keract.get_activations(model,ims)
#print("keys",act.keys())
#keract.display_activations(act)
'''
print("act",act.keys())
one_act = np.reshape(act["conv2d_1/BiasAdd:0"][0],(28,24,24))
print("shape",one_act.shape)
print(one_act[0].shape)
cv2.imshow("t",one_act[0], cv2.IMREAD_GRAYSCALE)
cv2.waitKey()
'''
y_preds_roc = [(x2-x1) for x1,x2 in y_preds]
fpr, tpr, thresholds = metrics.roc_curve(lbls,y_preds_roc)
#print("AUC",metrics.roc_auc_score(lbls,y_preds_roc))
#plt.plot(fpr,tpr)
def test_display_1(self):
model, x = dummy_model_and_inputs()
acts = get_activations(model, x)
display_activations(acts, save=True)
def test_compile_vgg16_model(self):
model, x = dummy_model_and_inputs()
model.name = 'vgg16' # spoof identity here!
get_activations(model, x, auto_compile=False)
self.assertTrue(model._is_compiled)
import sys
int_to_char = dict((i, c) for i, c in enumerate(chars))
start = np.random.randint(0, len(dataX)-1)
pattern = dataX[start]
print ("Seed:")
print ("\"", ''.join([int_to_char[value] for value in pattern]), "\"")
# generate characters
for i in range(100):
out_x = np.reshape(pattern, (1, len(pattern), 1))
out_x = out_x / float(n_vocab)
prediction = m.predict(out_x, verbose=0)
index = np.argmax(prediction)
result = int_to_char[index]
seq_in = [int_to_char[value] for value in pattern]
sys.stdout.write(result)
pattern.append(index)
pattern = pattern[1:len(pattern)]
print ("\nDone.")
num_simulations = x.shape[2]
attention_vectors = np.zeros(shape=(num_simulations, seq_length))
for i in range(num_simulations):
#testing_inputs_1, testing_outputs = get_data_recurrent(1, TIME_STEPS, INPUT_DIM)
activations = get_activations(m, x, layer_name='attention_weight')
#activations = K.function([m.layers[0].input], [m.layers[1].output])
attention_vec = np.mean(activations['attention_weight'], axis=0).squeeze()
assert np.abs(np.sum(attention_vec) - 1.0) < 1e-5
attention_vectors[i] = attention_vec
print("attention vector: ",attention_vectors)
# Section - 5
# Visualise the filters
conv_1 = model.get_layer('block1_conv1')
weights_1 = conv_1.get_weights()
print(weights_1[0].shape)
plt.figure(figsize=(6,3))
plt.subplot(1,2,1)
plt.imshow(normalize_image(weights_1[0][:, :, :, 0]))
plt.title('first layer - channel 1 weights')
plt.subplot(1,2,2)
plt.imshow(normalize_image(weights_1[0][:, :, :, 20]))
plt.title('first layer - channel 20 weights')
activations = get_activations(model, resident_owls_bogart_GEO)
# layer_name ='block1_conv1_1/Relu:0'
layer_name = list(activations.keys())[1]
print(layer_name)
show_activations(model, resident_owls_bogart_GEO, layer_name, 'GEO')
show_activations(model, resident_owls_bogart_COLOR, layer_name, 'COLOR')
show_activations(model, resident_owls_bogart_FILT, layer_name, 'FILT')
# Section - 6
#load the images
uploaded = files.upload() #please uploaad all the dogs images
uploaded = files.upload() #please uploaad all the cats images
!ls
import numpy as np
import requests
from PIL import Image
from keras.applications.vgg16 import VGG16
from keras.applications.vgg16 import decode_predictions
from keras.applications.vgg16 import preprocess_input
from keras.preprocessing.image import img_to_array
import keract
model = VGG16()
url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/1/14/Gatto_europeo4.jpg/250px-Gatto_europeo4.jpg'
response = requests.get(url)
image = Image.open(BytesIO(response.content))
image = image.crop((0, 0, 224, 224))
image = img_to_array(image)
arr_image = np.array(image)
image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
image = preprocess_input(image)
yhat = model.predict(image)
label = decode_predictions(yhat)
label = label[0][0]
print('{} ({})'.format(label[1], label[2] * 100))
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
activations = keract.get_activations(model, image, layer_name='block1_conv1')
keract.display_heatmaps(activations, arr_image, save=True)
model.fit(X_train, to_categorical(y_train), epochs=5, batch_size=64)
#model.evaluate(X_test,to_categorical(y_test))
predictions = model.predict(X_test)
def normalize(probs):
prob_factor = 1 / sum(probs)
return [prob_factor * p for p in probs]
n = 0
for x_test in X_test[:10]:
attention_map = get_activations(model,
np.array([x_test]),
layer_names='attention_weight')
a = attention_map['attention_weight'][0]
total = 0.0
for i in range(len(a)):
if i == focus_position - 1:
continue
total += a[i]
p_max = 0.0
for i in range(len(a)):
if i == focus_position - 1:
continue
a[i] /= total
if a[i] > p_max:
p_max = a[i]
a = np.array([a])
from tensorflow.keras.applications.vgg16 import VGG16
from tensorflow.keras.applications.vgg16 import decode_predictions
from tensorflow.keras.applications.vgg16 import preprocess_input
from tensorflow.keras.preprocessing.image import img_to_array
import keract
from utils import gpu_dynamic_mem_growth
if __name__ == "__main__":
# Check for GPUs and set them to dynamically grow memory as needed
# Avoids OOM from tensorflow greedily allocating GPU memory
gpu_dynamic_mem_growth()
model = VGG16()
image = Image.open('250px-Gatto_europeo4.jpeg')
image = image.crop((0, 0, 224, 224))
image = img_to_array(image)
image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
image = preprocess_input(image)
yhat = model.predict(image)
label = decode_predictions(yhat)
label = label[0][0]
print('{} ({})'.format(label[1], label[2] * 100))
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
activations = keract.get_activations(model, image)
first = activations.get('block1_conv1')
keract.display_activations(activations, save=True)
if __name__ == '__main__':
np.random.seed(123)
inp_a = np.random.uniform(size=(5, 10))
inp_b = np.random.uniform(size=(5, 10))
out_c = np.random.uniform(size=(5, 1))
# Just for visual purposes.
np.set_printoptions(precision=2)
# Activations of all the layers
print('MULTI-INPUT MODEL')
m1 = get_multi_inputs_model()
m1.compile(optimizer='adam', loss='mse')
utils.print_names_and_values(keract.get_activations(m1, [inp_a, inp_b]))
utils.print_names_and_values(
keract.get_gradients_of_trainable_weights(m1, [inp_a, inp_b], out_c))
utils.print_names_and_values(
keract.get_gradients_of_activations(m1, [inp_a, inp_b], out_c))
# Just get the last layer!
utils.print_names_and_values(
keract.get_activations(m1, [inp_a, inp_b], layer_name='last_layer'))
utils.print_names_and_values(
keract.get_gradients_of_activations(m1, [inp_a, inp_b],
out_c,
layer_name='last_layer'))
print('')
print('SINGLE-INPUT MODEL')
def test_nodes_empty(self):
model, x = dummy_model_and_inputs()
self.assertRaises(
ValueError,
lambda: get_activations(model, x, nodes_to_evaluate=[]))
[[255, 255, 255], [255, 255, 255], [255, 255, 255]]])
obj6 = np.array([[[0, 0, 0], [255, 255, 255], [0, 0, 0]],
[[255, 255, 255], [255, 255, 255], [255, 255, 255]]])
x = np.array([obj1, obj2, obj3, obj4, obj5, obj6])
x = x.reshape((1, x.size))
print(x.shape)
model = Sequential()
hidden = Dense(3, input_dim=4, activation='relu', name="hidden")
model.add(hidden)
model.add(Dense(4, name="output"))
model.compile(loss='mean_squared_error', optimizer='adam')
plot_model(model)
model.fit(x, x, epochs=1000)
a = get_activations(model, x)
b = list(a.values())
print(model.predict(np.array([[10, 10, 20, 20]])))
mean = np.mean(b[0])
std = np.std(b[0])
gauss = np.random.normal(mean, std, (1, 4))
model2 = Sequential()
model2.add(
Dense(4, input_dim=3, weights=model.get_layer("output").get_weights()))
print(model2.predict(list(a.values())[0]))
def test_compile_vgg16_model(self):
model, x = dummy_model_and_inputs(name="vgg16")
get_activations(model, x, auto_compile=False)
self.assertTrue(model._is_compiled)
i = i + 1
yield [x_batch_depth, y_batch_depth]
##### for multimodal
for image_embedding_rgb, image_embedding_depth, subject in test_generator_multimodal(
):
# activations = {}
# img_abs_path = os.path.join(image_dir_rgb, image)
# image_embedding = preprocess_image(img_abs_path)
# print(image_embedding)
# break
arr = np.array(
list(
get_activations(model_vgg_multimodal,
[image_embedding_rgb, image_embedding_depth],
layer_name).values())[0])
# arr_depth = np.array(list(get_activations(model_vgg_multimodal, image_embedding_depth,layer_name).values())[0])
all_image_list.append(arr)
subject_list.append(subject)
all_image_list = np.mean(np.array(all_image_list), axis=1)
subject_list = np.mean(np.array(subject_list), axis=1)
subject_list_label = np.where(subject_list == 1)[1]
#calculate tsne embeddings
X_tsne = TSNE(n_components=2).fit_transform(all_image_list)
dataset_x_tsne = pd.DataFrame({
't-SNE_Dim1': X_tsne[:, 0],