I wish to modify the CuDNNGRU Output with for loop. However, it seems like I can't do so due to the tf.GradientTape graph mode. How can I modify the CuDNNGRU in Functional API? I know that normally we can perform some matrix ops on the functional API with tf.keras.backend.* functions like K.backend.batch_dot etc. However, I have to do some complex ops like triple for loop or more, etc., If someone knows how to do so, please help!
.....source code
x = L.Lambda(lambda fm: tf.squeeze(fm, axis=1))(x)
gru_1 = CuDNNGRU(512, return_sequences=True, name='gru1')(x)
gru_1b = CuDNNGRU(512, return_sequences=True, go_backwards=True,name='gru1_b')(x)
for i in gru_1:
.....apply some function to gru_1 outputs
By the way, I currently try to modify the GRU outputs with below code.
def attention(inputs):
transpose_input = tf.transpose(inputs,perm=[0,2,1])
atten_w = K.backend.batch_dot(inputs,transpose_input)
atten_w = tf.linalg.set_diag(atten_w,tf.zeros(tf.shape(atten_w)[0:-1],dtype=tf.float32))
atten_w = tf.nn.softmax(atten_w,axis=1)
atten_v = tf.py_function(calculate_atten,inp=[inputs,atten_w],Tout=[tf.float32])
atten_v = tf.convert_to_tensor(atten_v)
atten_v.set_shape(self.input_shapex)
def calculate_atten(data,atten_w):
input_vector = data.numpy()
atten_vectors = atten_w.numpy()
all_batch = []
for index,one_batch in enumerate(input_vector):
tmp_w = atten_vectors[index]
all_vector = []
for j,vector in enumerate(one_batch):
tmp = np.zeros(input_vector.shape[2])
for w in tmp_w[j]:
tmp += vector*w
all_vector.append(tmp)
all_batch.append(all_vector)
return all_batch
However, the above code, the tf.py_function return [time,features] instead of [batch,time,features], if this can be done, i can use tf.py_function to build a layer. But it seems like can't, HELP!!!!
I have been able to achieve the ops, with nested tf.map_fn. Although it needs to think properly what to pass to tf.map_fn(multiple inputs must be return in multiple output). Hope this can help others
def attn_transformation(inputs):
inputs_transpose = tf.transpose(inputs)
atten_w = tf.tensordot(inputs,inputs_transpose,axes=1)
def transform(data):
multiply_data = data[0]*data[1][...,tf.newaxis]
return [multiply_data,data[1]]
data = tf.map_fn(lambda x:transform(x),elems=([inputs,atten_w]))
data = tf.reduce_sum(data[0],axis=1)
return data
gru_1 = CuDNNGRU(512, return_sequences=True, name='gru1')(x)
gru_1b = CuDNNGRU(512, return_sequences=True, go_backwards=True,name='gru1_b')(x)
atten_vf = L.Lambda(lambda x: tf.map_fn(attn_transformation,x))(gru_1)
For any arbitrary operation where you want to apply it to every i
in tensor
you can just use tf.map_fn()
So for example, we can do something like:
inp = Input(shape=(2,3))
gru = CuDNNGRU(512, return_sequences=True)(inp)
def dummy_operation_to_be_applied(row):
return row + 1
out = Lambda(lambda x: tf.map_fn(dummy_operation_to_be_applied, x))(gru)
UPDATE:
Note that we can also nest tf.map_fn()
to map operations across lower dimensions too.
For example:
def nested_op(x):
return tf.reduce_max(x) + x
def dummy_operation_to_be_applied(row):
return tf.map_fn(nested_op, row)
out = Lambda(lambda x: tf.map_fn(dummy_operation_to_be_applied, x))(gru)
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.