İ try to use tensorflow image retraining.
https://www.tensorflow.org/tutorials/image_retraining
train like that and it is OK:
D:\dev\Anaconda\python D:/dev/detect_objects/tensorflow-master/tensorflow/examples/image_retraining/retrain.py --image_dir D:/dev/detect_objects/flower_photos --bottleneck_dir D:/dev/detect_objects/tensorflow-master/retrain/bottleneck --architecture mobilenet_0.25_128 --output_graph D:/dev/detect_objects/tensorflow-master/retrain/output_graph/output.pb --output_labels D:/dev/detect_objects/tensorflow-master/retrain/output_labels/labels.txt --saved_model_dir D:/dev/detect_objects/tensorflow-master/retrain/saved_model_dir --how_many_training_steps 100
When predict new image like:
D:\dev\Anaconda\python D:/dev/detect_objects/tensorflow-master/tensorflow/examples/label_image/label_image.py --graph=D:/dev/detect_objects/tensorflow-master/retrain/output_graph/output.pb --labels=D:/dev/detect_objects/tensorflow-master/retrain/output_labels/labels.txt --image=D:/dev/detect_objects/flower_photos/daisy/21652746_cc379e0eea_m.jpg
It gives error
KeyError: "The name 'import/Mul' refers to an Operation not in the graph."
label_image.py content:
input_height = 299
input_width = 299
input_mean = 0
input_std = 255
#input_layer = "input"
#output_layer = "InceptionV3/Predictions/Reshape_1"
input_layer = "Mul"
output_layer = "final_result"
What is the problem here?
Change this:
input_height = 299
input_width = 299
input_mean = 0
input_std = 255
#input_layer = "input"
#output_layer = "InceptionV3/Predictions/Reshape_1"
input_layer = "Mul"
output_layer = "final_result"
to this:
input_height = 128
input_width = 128
input_mean = 0
input_std = 128
input_layer = "input"
output_layer = "final_result"
If there is no node in the graph called "import/Mul" and we don't know
what the graph is or how it was produced, there's little chance that
anyone will be able to guess the right answer.
You might try printing the list of operations of your graph using graph.get_operations() and attempting to locate an appropriate-sounding node (try the first one that is printed)
Related
I want use Instance Normalization in a converted model (Keras to CoreML).
Keras contrib Instance Normalization is treated as custom layer in coreML. So I copied custom layer like this.
def create_instance_normalization_spec(layer):
input_name = layer._inbound_nodes[0].inbound_layers[0].name
input_name += '_output'
output_name = layer.name + '_output'
spec_layer = NeuralNetwork_pb2.NeuralNetworkLayer()
spec_layer.name = layer.name
spec_layer.input.append(input_name)
spec_layer.output.append(output_name)
spec_layer_params = spec_layer.batchnorm
weights = layer.get_weights()
channels = weights[0].shape[0]
idx = 0
gamma, beta = None, None
if layer.scale:
gamma = weights[idx]
idx += 1
if layer.center:
beta = weights[idx]
idx += 1
epsilon = layer.epsilon or 1e-5
spec_layer_params.channels = channels
spec_layer_params.gamma.floatValue.extend(map(float,
gamma.flatten()))
spec_layer_params.beta.floatValue.extend(map(float,
beta.flatten()))
spec_layer_params.epsilon = epsilon
spec_layer_params.computeMeanVar = True
spec_layer_params.instanceNormalization = True
instance_norm_spec =
create_instance_normalization_spec(keras_model.layers[-1])
instance_norm_spec.input[:] = ["input1"]
instance_norm_spec.output[:] = ["output1"]
mlmodel._spec.neuralNetwork.layers[-1].CopyFrom(instance_norm_spec)
Now, this coreML model can predict.But outputs are wrong.
Is the way above wrong? How can I fix it?
from keras import regularizers,constraints,initializers
model = Sequential()
FIRST_LSTM = 256
SECOND_LSTM = 256
EMBEDD_VECTOR_LENGTH = 128
outputs=[]
inputs_=[]
EMB = Embedding(DICTIONARY_LENGTH, EMBEDD_VECTOR_LENGTH, input_length=30)
BDR = Bidirectional(CuDNNLSTM(FIRST_LSTM,return_sequences=True),merge_mode='concat')
TDB = TimeDistributed(Dense(DICTIONARY_LENGTH+3))
for i in range(SENTENCES_IN_NOVEL): # SENTENCES_IN_NOVEL = 100
if(i%10==0):
print(i,"/",SENTENCES_IN_NOVEL)
inputlayer = Input(shape=[MAX_LENGTH]) # MAX_LENGTH = 30 (30 words per sentence)
inputs_.append(inputlayer)
layer = EMB(inputlayer) # 50 is embedd vector size for each POS Tag
layer = BDR(layer) # output 100 because of bidirectonal concat
layer = Lambda( lambda x: K.sum(x, axis=1), input_shape=(30,2*FIRST_LSTM))(layer)
outputs.append(layer)
merge_ = concatenate(outputs)
merge_ = Reshape((SENTENCES_IN_NOVEL, 2*FIRST_LSTM), input_shape=(SENTENCES_IN_NOVEL*(2*FIRST_LSTM),))(merge_)
merge_ = Bidirectional(CuDNNLSTM(SECOND_LSTM,return_sequences=True),merge_mode='concat')(merge_)
attention_mul = Attention(100)(merge_)#step_dimen =100
output = Dense(14, activation='softmax')(attention_mul)
model = Model(inputs=inputs_,outputs=output)
model.summary()
I wrote this model to predict author of Novels LSTM-LSTM Model.
Wants to convert CNN-LSTM.
I am unable to figure out this problem.
Trying to recreate the Tensorflow NMT example
https://github.com/tensorflow/nmt#bidirectional-rnns
with a baseline of just copying the original sentence instead of translating, but getting this weird bug for my inference part where the only outputs are the same words multiple times.
Using the TF attention wrapper and using Greedy Embedding Helper for inference.
Using TF 1.3 and python 3.6 if that helps
Screenshot of the bugged prediction
The weird thing is during training, the predictions are normal and the loss decreased to around 0.1
I have already checked the embeddings and they do change from each time step and I suspect that this has something to do with the decoding stage since it's the only part the really changes from training to inference.
tf.reset_default_graph()
sess = tf.InteractiveSession()
PAD = 0
EOS = 1
max_gradient_norm = 1
learning_rate = 0.02
num_layers = 1
total_epoch = 2
sentence_length = 19
vocab_size = 26236
input_embedding_size = 128
if mode == "training":
batch_size = 100
isReused = None
else:
batch_size = 1
isReused = True
with tf.name_scope("encoder"):
encoder_embeddings = tf.get_variable('encoder_embeddings', [vocab_size, input_embedding_size], tf.float32,
tf.random_uniform_initializer(-1.0, 1.0))
encoder_hidden_units = 128
encoder_inputs = tf.placeholder(shape=(batch_size, None), dtype=tf.int32, name='encoder_inputs')
encoder_lengths = tf.placeholder(shape=batch_size, dtype=tf.int32, name='encoder_lengths')
encoder_cell = tf.contrib.rnn.BasicLSTMCell(encoder_hidden_units, state_is_tuple=True)
encoder_inputs_embedded = tf.nn.embedding_lookup(encoder_embeddings, encoder_inputs)
encoder_outputs, encoder_state = tf.nn.dynamic_rnn(encoder_cell, encoder_inputs_embedded, dtype=tf.float32,
sequence_length=encoder_lengths, time_major=False)
with tf.variable_scope("decoder", isReused):
decoder_hidden_units = encoder_hidden_units
decoder_inputs = tf.placeholder(shape=(batch_size, None), dtype=tf.int32, name='decoder_inputs')
decoder_targets = tf.placeholder(shape=(batch_size, None), dtype=tf.int32, name='decoder_targets')
decoder_lengths = tf.placeholder(shape=batch_size, dtype=tf.int32, name="decoder_lengths")
decoder_embeddings = tf.get_variable('decoder_embeddings', [vocab_size, input_embedding_size], tf.float32,
tf.random_uniform_initializer(-1.0, 1.0))
decoder_inputs_embedded = tf.nn.embedding_lookup(decoder_embeddings, decoder_inputs)
decoder_cell = tf.contrib.rnn.BasicLSTMCell(decoder_hidden_units, state_is_tuple=True)
projection_layer = layers_core.Dense(vocab_size, use_bias=False)
attention_mechanism = tf.contrib.seq2seq.LuongAttention(encoder_hidden_units, encoder_outputs,
memory_sequence_length=encoder_lengths)
attn_decoder_cell = tf.contrib.seq2seq.AttentionWrapper(decoder_cell, attention_mechanism,
attention_layer_size=encoder_hidden_units)
if mode == "training":
helper = tf.contrib.seq2seq.TrainingHelper(decoder_inputs_embedded, decoder_lengths, time_major=False)
maximum_iterations = None
else:
helper = tf.contrib.seq2seq.GreedyEmbeddingHelper(decoder_embeddings, tf.fill([batch_size], EOS), EOS)
maximum_iterations = tf.round(tf.reduce_max(encoder_lengths) * 2)
# Decoder
init_state = attn_decoder_cell.zero_state(batch_size, tf.float32).clone(cell_state=encoder_state)
decoder = tf.contrib.seq2seq.BasicDecoder(attn_decoder_cell, helper, init_state, output_layer=projection_layer)
# Dynamic decoding
decoder_outputs, decoder_final_state, _ = tf.contrib.seq2seq.dynamic_decode(decoder, output_time_major=False,
swap_memory=True,
maximum_iterations=maximum_iterations)
decoder_logits = decoder_outputs.rnn_output
decoder_prediction = decoder_outputs.sample_id
if mode == "training":
with tf.name_scope("cross_entropy"):
labels = tf.one_hot(decoder_targets, depth=vocab_size, dtype=tf.float32)
decoder_crossent = tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=decoder_logits)
with tf.name_scope("loss"):
target_weights = tf.sequence_mask(decoder_lengths, maxlen=20, dtype=decoder_logits.dtype)
train_loss = tf.reduce_sum(decoder_crossent * target_weights) / (batch_size * 20)
tf.summary.scalar('loss', train_loss)
with tf.name_scope("clip_gradients"):
params = tf.trainable_variables()
gradients = tf.gradients(train_loss, params)
clipped_gradients, _ = tf.clip_by_global_norm(gradients, max_gradient_norm)
with tf.name_scope("Optimizer"):
optimizer = tf.train.AdamOptimizer(learning_rate)
update_step = optimizer.apply_gradients(zip(clipped_gradients, params))
merged = tf.summary.merge_all()
train_writer = tf.summary.FileWriter(os.getcwd() + '/train', sess.graph)
test_writer = tf.summary.FileWriter(os.getcwd() + '/test', )
tf.global_variables_initializer().run()
sess.run(tf.global_variables_initializer())
I am using libsvm java package for a sentence classification task. I have 3 classes. Every sentence is represented as a vector of size 435. The format of vector_file is as follows:
1 0 0.12 0 0.5 0.24 0.32 0 0 0 ... 0.43 0 First digit indicates class label and remaining is the vector.
The following is how I am making the svm_problem:
public void makeSvmProb(ArrayList<Float> inputVector,float label,int p){
// p is 0 to 77 (total training sentences)
int idx=0,count=0;
svm_prob.y[p]=label;
for(int i=0;i<inputVector.size();i++){
if(inputVector.get(i)!=0) {
count++; // To get the count of non-zero values
}
}
svm_node[] x = new svm_node[count];
for(int i=0;i<inputVector.size();i++){
if(inputVector.get(i)!=0){
x[idx] = new svm_node();
x[idx].index = i;
x[idx].value = inputVector.get(i);
idx++;
}
}
svm_prob.x[p]=x;
}
Parameter settings:
param.svm_type = svm_parameter.C_SVC;
param.kernel_type = svm_parameter.RBF;
param.degree = 3;
param.gamma = 0.5;
param.coef0 = 0;
param.nu = 0.5;
param.cache_size = 40;
param.C = 1;
param.eps = 1e-3;
param.p = 0.1;
param.shrinking = 1;
param.probability = 0;
param.nr_weight = 0;
param.weight_label = new int[0];
param.weight = new double[0];
While executing the program, After 2 iterations, I am getting a NullPointerException. I couldn't figure out what is going wrong.
This is the error coming:
optimization finished, #iter = 85
nu = 0.07502654779820772
obj = -15.305162227093849, rho = -0.03157808477381625
nSV = 47, nBSV = 1
*
optimization finished, #iter = 88
nu = 0.08576821199868506
obj = -17.83925196551639, rho = 0.1297986754900152
nSV = 51, nBSV = 3
Exception in thread "main" java.lang.NullPointerException
at libsvm.Kernel.dot(svm.java:207)
at libsvm.Kernel.<init>(svm.java:199)
at libsvm.SVC_Q.<init>(svm.java:1156)
at libsvm.svm.solve_c_svc(svm.java:1333)
at libsvm.svm.svm_train_one(svm.java:1510)
at libsvm.svm.svm_train(svm.java:2067)
at SvmOp.<init>(SvmOp.java:130)
at Main.main(Main.java:8)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at com.intellij.rt.execution.application.AppMain.main(AppMain.java:147)
Any idea on what is going wrong?
The NullPointerException is thrown in Line 207 in svm.class. Investigating the source code shows:
static double dot(svm_node[] x, svm_node[] y)
{
double sum = 0;
int xlen = x.length;
...
}
Line 207 is int xlen = x.length;. So in this case, we see, that one of your svm_node (or vectors) is null.
For this reason, we cannot really help you here, as we would need more information / source code to debug it.
I would go for the following strategy:
Investigate the svm_node objects after you completed the building of the svm_problem in a debugger and look for null values.
Check the build process of your svm_problem. The problem might be there.
An other possibility would be to change your data-format and be compliant to the official LIBSVM format:
As stated in the documentation, the data format uses sparse-data format and should be like that:
<label> 0:i 1:K(xi,x1) ... L:K(xi,xL)
The ascending integer refers to the attribute or feature id, which is necessary for internal representation of the vector.
I previously replied to a similar question here and added an example for the data format.
This format can be read out of the box as the code to construct the svm_problem is included in the library.
I keep getting this error when I try to compile. I am not quite sure what the issue is. I am not the best at coding, and thus, debugging is not my strong suit. Any ideas whats causing the error?
PROGRAM Atmospheric_Reflection
IMPLICIT NONE
REAL*8:: Wavelength = 0.200D-6, WaveNumber = 0.0, pi = 3.14159265, Length(1:71) = 0, Number(1:71) = 0, IndexOfRe(1:71) = 0
INTEGER:: i = 0
!Calculate the wave number for wavelengths varying from 0.2 micrometers to 1.6 micrometers in increments of 0.02 micrometers
i = 1
DO WHILE (Wavelength <= 1.600D-6)
WaveNumber = 1/Wavelength
Length(i) = Wavelength
Number(i) = WaveNumber
IndexOfRe(i) = 1 + (((5791817.0/(238.0185 - WaveNumber**2))+(167909.0/(57.362-WaveNumber**2))))D-8
Wavelength = Wavelength + 0.02D-6
i = i+1
END DO
END PROGRAM
Possibly the D-8 at the end of this line
IndexOfRe(i) = 1 + (((5791817.0/(238.0185 - WaveNumber**2))+(167909.0/(57.362-WaveNumber**2))))D-8