So, my code is like
parsed_line = tf.decode_csv(line, [[0], [0], [""]])
print(parsed_line[0])
del parsed_line[0]
del parsed_line[0]
features = parsed_line
print(parsed_line[0])
then the result is
[<tf.Tensor 'DecodeCSV:0' shape=() dtype=int32>, <tf.Tensor 'DecodeCSV:1' shape=() dtype=int32>, <tf.Tensor 'DecodeCSV:2' shape=() dtype=string>]
and
[<Tensor("DecodeCSV:2", shape=(), dtype=string)>]
the csv i will give to this decode function is
1, 0, 0101010010101010101010
and I want this "0101010010101010101010" to
[0,1,0,1,0,.........]
in tensorflow
[<Tensor("DecodeCSV:2", shape=(), dtype=string)>]
to
[<tf.Tensor 'DecodeCSV:0' shape=() dtype=int32>, <tf.Tensor 'DecodeCSV:1' shape=() dtype=int32>, ............]
do you have any ideas of this?
You can do it this way using tf.string_split and tf.string_to_number:
import tensorflow as tf
line = tf.constant("1000111101", shape=(1,))
b = tf.string_split(line, delimiter="").values
c = tf.string_to_number(b, tf.int32)
print(c)
with tf.Session() as sess:
print(sess.run(c))
[1 0 0 0 1 1 1 1 0 1]
Related
I try to understand how to add trainable variable in my model.
I wrote some code :
def custom_loss( y_true, y_pred, k1, k2, k3):
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=y_true, logits=y_pred))
cosine0 = cosine_dist(y_true[:MAX_BITS],y_pred[:MAX_BITS])
error = custom___error(y_true,y_pred)
loss = loss * k1 + cosine0 * k2 + k3 * error
return loss
model.compile(loss=lambda y_true, y_pred: custom_loss(y_true, y_pred, k1, k2, k3), optimizer=optimizer, metrics=[‘accuracy’,custom_error,custom___error])
But the variables k1, k2, k3 never change : how can I add them to the trainable variables list ?
k1: <tf.Variable ‘Variable:0’ shape=() dtype=float32, numpy=0.5>
k2: <tf.Variable ‘Variable:0’ shape=() dtype=float32, numpy=0.5>
k3: <tf.Variable ‘Variable:0’ shape=() dtype=float32, numpy=0.5>
Regards
I trained my customized ssd_mobilenet_v2 using TensorFlow2 Object Detection API.
After training completed, I used exporter_main_v2.py to export a saved_model of my customized model.
If I load saved_model by TensorFlow2, it seem there are two kind of output format.
import tensorflow as tf
saved_model = tf.saved_model.load("saved_model")
detect_fn = saved_model["serving_default"]
print(detect_fn.outputs)
'''
[<tf.Tensor 'Identity:0' shape=(1, 100) dtype=float32>,
<tf.Tensor 'Identity_1:0' shape=(1, 100, 4) dtype=float32>,
<tf.Tensor 'Identity_2:0' shape=(1, 100) dtype=float32>,
<tf.Tensor 'Identity_3:0' shape=(1, 100, 7) dtype=float32>,
<tf.Tensor 'Identity_4:0' shape=(1, 100) dtype=float32>,
<tf.Tensor 'Identity_5:0' shape=(1,) dtype=float32>,
<tf.Tensor 'Identity_6:0' shape=(1, 1917, 4) dtype=float32>,
<tf.Tensor 'Identity_7:0' shape=(1, 1917, 7) dtype=float32>]
'''
print(detect_fn.structured_outputs)
'''
{'detection_classes': TensorSpec(shape=(1, 100), dtype=tf.float32, name='detection_classes'),
'detection_scores': TensorSpec(shape=(1, 100), dtype=tf.float32, name='detection_scores'),
'detection_multiclass_scores': TensorSpec(shape=(1, 100, 7), dtype=tf.float32, name='detection_multiclass_scores'),
'num_detections': TensorSpec(shape=(1,), dtype=tf.float32, name='num_detections'),
'raw_detection_boxes': TensorSpec(shape=(1, 1917, 4), dtype=tf.float32, name='raw_detection_boxes'),
'detection_boxes': TensorSpec(shape=(1, 100, 4), dtype=tf.float32, name='detection_boxes'),
'detection_anchor_indices': TensorSpec(shape=(1, 100), dtype=tf.float32, name='detection_anchor_indices'),
'raw_detection_scores': TensorSpec(shape=(1, 1917, 7), dtype=tf.float32, name='raw_detection_scores')}
'''
Then, I try to convert this saved_model to onnx format using tf2onnx.
However, the outputs of onnxruntime was a list.
By the shape of result in the list, I think that the sequence is same as detect_fn.outputs
import numpy as np
import onnxruntime as rt
sess = rt.InferenceSession("model.onnx")
input_name = sess.get_inputs()[0].name
pred_onx = sess.run(None, {input_name: np.zeros((1,300,300,3), dtype=np.uint8)})
print(pred_onx) # a list
print([i.shape for i in pred_onx])
'''
[(1, 100),
(1, 100, 4),
(1, 100),
(1, 100, 7),
(1, 100),
(1,),
(1, 1917, 4),
(1, 1917, 7)]
'''
Because there is some shape of result which is same as others, so it become hard to recognized.
Is there any document talk about this relationship that I can refer?
After I looked closely into the values in outputs.
I found the mapping relationship below.
def result_mapper(outputs):
result_dict = dict()
result_dict["num_detections"] = outputs[5]
result_dict["raw_detection_scores"] = outputs[7]
result_dict["raw_detection_boxes"] = outputs[6]
result_dict["detection_multiclass_scores"] = outputs[3]
result_dict["detection_boxes"] = outputs[1]
result_dict["detection_scores"] = outputs[4]
result_dict["detection_classes"] = outputs[2]
result_dict["detection_anchor_indices"] = outputs[0]
return result_dict
Had the same question and after some hours of stepping through the debugger found they are... in sorted order. The method determining iteration order is here.
In the case of dict instances, the sequence consists of the values, sorted by
key to ensure deterministic behavior.
I can get a tensor by graph.get_tensor_by_name, however I cannot find it in tf.global_variable.
In my case, I defined some tf.Tensor as follows:
output_y = Dense(units=y.shape[1],activation='softmax',kernel_regularizer=regularizers.l2(),bias_regularizer=regularizers.l2(),activity_regularizer=regularizers.l2(),name='output_y_'+str(index))(pretrain_output)
y_tf = tf.placeholder(tf.float32, shape=(None, y.shape[1]),name='y_tf_'+str(index))
loss_tensor = tf.nn.softmax_cross_entropy_with_logits(logits=output_y, labels=y_tf, name='loss_tensor_' + str(index))
I can export the tensor shape and name as follows:
>>output_y
<tf.Tensor 'train_variable/output_y_0/Softmax:0' shape=(?, 4) dtype=float32>
>>y_tf
<tf.Tensor 'train_variable/y_tf_0:0' shape=(?, 4) dtype=float32>
>>loss_tensor
<tf.Tensor 'train_variable/loss_tensor_0/Reshape_2:0' shape=(?,) dtype=float32>
Also, I can use tf.get_default_graph.get_tensor_by_name to retrieve the tensor:
>>tf.get_default_graph().get_tensor_by_name('train_variable/output_y_0/Softmax:0')
<tf.Tensor 'train_variable/output_y_0/Softmax:0' shape=(?, 4) dtype=float32>
>>tf.get_default_graph().get_tensor_by_name('train_variable/y_tf_0:0')
<tf.Tensor 'train_variable/y_tf_0:0' shape=(?, 4) dtype=float32>
>>tf.get_default_graph().get_tensor_by_name('train_variable/loss_tensor_0/Reshape_2:0')
<tf.Tensor 'train_variable/loss_tensor_0/Reshape_2:0' shape=(?,) dtype=float32>
However, these variable names cannot be found in tf.global_variables(). It seems that tf.global_variables() only contains the parameter variables like kernel/bias. Now I have to remember the tensor name in order to retrieve the object output (output_y in my case). Can someone show me how to retrieve a tensor, for example search it in a list with all tensor?
There is a difference between a tensor from the read-operation of a node and a tensor as a variable.
A variable consists of a value and several operations:
import tensorflow as tf
a = tf.get_variable('a', tf.float32)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
sess.run(a) # gives 42.
sess.run(tf.get_default_graph().get_tensor_by_name('a/read:0')) # gives 42. as well
print(a.op.outputs) # <tf.Tensor 'a:0' shape=() dtype=float32_ref>]
It just behaves similarly:
>>> type(a)
<class 'tensorflow.python.ops.variables.Variable'>
>>> type(tf.get_default_graph().get_tensor_by_name('a/read:0'))
<class 'tensorflow.python.framework.ops.Tensor'>
but they are different though.
The easiest way is to return output_y in case you need it again. Otherwise just follow:
https://stackoverflow.com/a/36893840/7443104
I have no idea how this error is arising. I'm trying to change the input format to an RNN and have printed out the tensors in the original version (which works) and the modified version (which crashes).
FUNCTIONAL:
LABEL= Tensor("concat_1:0", shape=(?, 2), dtype=float32, device=/device:CPU:0) (?, 2)
inputs=Tensor("concat:0", shape=(?, 8), dtype=float32, device=/device:CPU:0)
x=[<tf.Tensor 'split:0' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:1' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:2' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:3' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:4' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:5' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:6' shape=(?, 1) dtype=float32>,
<tf.Tensor 'split:7' shape=(?, 1) dtype=float32>]
last outputs=Tensor("rnn/rnn/basic_lstm_cell/mul_23:0", shape=(?, 3), dtype=float32)
PREDICTION Tensor("add:0", shape=(?, 2), dtype=float32)
LOSS Tensor("mean_squared_error/value:0", shape=(), dtype=float32)
BROKEN:
X= 5 Tensor("Const:0", shape=(49, 10), dtype=float32, device=/device:CPU:0)
labels= Tensor("Const_5:0", shape=(49, 10), dtype=float32)
OUTPUTS Tensor("rnn/rnn/basic_lstm_cell/mul_14:0", shape=(49, 5), dtype=float32)
PREDICTIONS Tensor("add:0", shape=(49, 10), dtype=float32)
LABELS Tensor("Const_5:0", shape=(49, 10), dtype=float32)
LOSS Tensor("mean_squared_error/value:0", shape=(), dtype=float32)
Here is the code for the model, which is the same for each of them:
lstm_cell = rnn.BasicLSTMCell(LSTM_SIZE, forget_bias=1.0)
outputs, _ = tf.nn.static_rnn(lstm_cell, x, dtype=tf.float32)
outputs = outputs[-1]
print('-->OUTPUTS', outputs)
weight = tf.Variable(tf.random_normal([LSTM_SIZE, N_OUTPUTS]))
bias = tf.Variable(tf.random_normal([N_OUTPUTS]))
predictions = tf.matmul(outputs, weight) + bias
print('-->PREDICTIONS', predictions)
print('-->LABELS', labels)
loss = tf.losses.mean_squared_error(labels, predictions)
print('-->LOSS', loss)
train_op = tf.contrib.layers.optimize_loss(loss=loss, global_step=tf.train.get_global_step(), learning_rate=0.01, optimizer="SGD")
eval_metric_ops = {"rmse": tf.metrics.root_mean_squared_error(labels, predictions)}
TL;DR: Use x = tf.split( x, 10, axis = -1 ) to split x before feeding it.
TS;WM:
The error probably happens at tf.nn_static_rnn(), in the second line of your code (would have been nice had you posted the error line number):
outputs, _ = tf.nn.static_rnn(lstm_cell, x, dtype=tf.float32)
The "broken" version tries to feed a tensor with shape ( 49, 10 ) whereas the working version is feeding a list of 8 tensors with shape ( ?, 1 ). The documentation says:
inputs: A length T list of inputs, each a Tensor of shape [batch_size, input_size], or a nested tuple of such elements.
In the previous line you define the lstm_cell with tf.contrib.rnn.BasicLSTMCell.__init__() (presumably, because the import lines are omitted from your code), which has the num_units argument filled by LSTM_SIZE (which is, again, omitted from your code):
lstm_cell = rnn.BasicLSTMCell(LSTM_SIZE, forget_bias=1.0)
So you have to get your ducks in a row. x has to be a list of ( batch_size, 1 ) tensors, which you can achieve with tf.split():
x = tf.split( x, 10, axis = -1 )
where I presume 10 to be the length of the data you're trying to feed, just based on the output you pasted.
Below code creates some tensors, which are implicit. I wonder how I can view values of these tensors:
<tf.Variable 'rnn/basic_lstm_cell/kernel:0' shape=(43, 160) dtype=float32_ref>
<tf.Variable 'rnn/basic_lstm_cell/bias:0' shape=(160,) dtype=float32_ref>
<tf.Variable 'rnn/basic_lstm_cell/kernel/Adagrad:0' shape=(43, 160) dtype=float32_ref>
<tf.Variable 'rnn/basic_lstm_cell/bias/Adagrad:0' shape=(160,) dtype=float32_ref>
<tf.Variable 'softmax/W/Adagrad:0' shape=(40, 10) dtype=float32_ref>
<tf.Variable 'softmax/b/Adagrad:0' shape=(10,) dtype=float32_ref>
Here is code itself.
import tensorflow as tf
import numpy as np
VECTOR_SIZE = 3
SEQUENCE_LENGTH = 5
BATCH_SIZE = 7
STATE_SIZE = 40
NUM_CLASSES = 10
LEARNING_RATE = 0.1
x = tf.placeholder(tf.float32, [BATCH_SIZE, SEQUENCE_LENGTH, VECTOR_SIZE],
name='input_placeholder')
y = tf.placeholder(tf.int32, [BATCH_SIZE, SEQUENCE_LENGTH],
name='labels_placeholder')
init_state = tf.zeros([BATCH_SIZE, STATE_SIZE])
rnn_inputs = tf.unstack(x, axis = 1)
y_as_list = tf.unstack(y, axis=1)
cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs,
initial_state=(init_state,init_state))
with tf.variable_scope('softmax'):
W = tf.get_variable('W', [STATE_SIZE, NUM_CLASSES])
b = tf.get_variable('b', [NUM_CLASSES], initializer=tf.constant_initializer(0.0))
logits = [tf.matmul(rnn_output, W) + b for rnn_output in rnn_outputs]
predictions = [tf.nn.softmax(logit) for logit in logits]
losses = [tf.nn.sparse_softmax_cross_entropy_with_logits(labels=label, logits=logit) for \
logit, label in zip(logits, y_as_list)]
total_loss = tf.reduce_mean(losses)
train_step = tf.train.AdagradOptimizer(LEARNING_RATE).minimize(total_loss)
X = np.ones([BATCH_SIZE, SEQUENCE_LENGTH, VECTOR_SIZE])
Y = np.ones([BATCH_SIZE, SEQUENCE_LENGTH])
saver = tf.train.Saver()
sess = tf.Session()
sess.run(tf.global_variables_initializer())
batch_total_loss = sess.run([total_loss, train_step],
feed_dict = {x:X,y:Y})
save_path = saver.save(sess, "/tmp/model.ckpt")
for el in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES):
print(el)
Use sess.run():
for el in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES):
print(el) # this will print the tensor's name, shape, data type
print(sess.run(el)) # this will print the tensor's current value