Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
Repo info
    Akshay Mankar
    @fuzzylogic666 you can try with tensorflow/TF by sentdex on youtube
    Ethan Hickman
    Hi all, I was wondering if anyone has dealt with what I'm trying to do before. My goal is to take the model I've trained in Keras with the Tensorflow backend and optimize the graph for inference. So far I've been trying this with MNIST hello world example. The problem I'm having is that when I get the graph from the keras backend session, the input node has a -1 where the 'None' was (which should be there to represent a wildcard value for batch size), but that keeps the graph from being rebuilt properly because the first layer has -1 in its placeholder dimensions! I get the following error.
    2017-07-07 17:25:53.977324: W tensorflow/core/framework/op_kernel.cc:1148] Invalid argument: Shape [-1,28,28,1] has negative dimensions
    2017-07-07 17:25:53.977393: E tensorflow/core/common_runtime/executor.cc:644] Executor failed to create kernel. Invalid argument: Shape [-1,28,28,1] has negative dimensions
         [[Node: input_1 = Placeholder[dtype=DT_FLOAT, shape=[?,28,28,1], _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
    Ethan Hickman
    If anyone comes up with an answer, this stack overflow question is asking pretty much the same thing: https://stackoverflow.com/questions/44930754/restore-a-model-trained-with-variable-input-length-in-tensorflow-results-in-inva
    Simon Ho
    May I know the difference between tf.estimator.EstimatorSpec and tf.contrib.learn.ModelFnOps?
    Jaffer Wilson
    May I know is there a very large set for tensorfow summarization? with all the training bin and vocab file?
    Can anyone help me with that?
    hi, all! i walked through this tutorial https://www.tensorflow.org/tutorials/wide and i have graph.pbtxt in output dir. But i need grapn in pb format, to use in TensorFlowSharp. Please, provide me with any way to get it.
    Anybody compiled static libraries (libtensorflow.a) for Android ?
    Trevor R.H. Clarke
    trying to write an input pipeline for some data in tf 1.2 python. I have a file format that has 3 bytes (number as string) followed by PNG data (variable length). I'm using the cifar10 tutorial as a starting point.
    I have a filename queue which I put into a wholefilereader and .read() to get the filename, raw_data. I do tf.string_to_number(tf.substr(raw_data, 0, 3), tf.int32) to get the numeric value (it's the class number for the image)
    having trouble getting the png data. tried tf.image.decode_png(tf.substr(raw_data, 3, -1)) and it said the -1 wasn't a valid index. Next I tried tf.image.decode_png(tf.substr(raw_data, 3, string_length(raw_data)) where string_length is
    def string_length(t): return tf.py_func(lambda p: [len(x) for x in p], [t], [tf.int32])[0]
    with that I get ValueError: Shape must be rank -1 but is rank 0 for 'Substr' (op: 'Substr') with input shapes: [], [], ?.
    I'm guessing that might mean that my rank 0 int32 tensor can't be used? I don't want to pass in a session and eval() the string_length result forcing the whole input pipe to eval at that point. How else can I do this?
    Shivam Kalra
    Can anyone explain how does conv2d work on multi-channel image?
    Is convolution calculated per channel and then added to each other per filter?
    Simon Ho
    Is there any way to implement Multi-task learning with Estimator ?
    i.e. multiple outputs with model_fn
    Sebastian Raschka
    Shivam, regarding the conv2d. E.g. say we have 5x5 image with 3 channels. I.e., the image is 5x5x3. Now, you have 2 kernels. Each kernel has the size 3x3x3. The first kernel will do the convolution for each RGB channel, adding the results together and give you a 3x3 feature map (if you don't consider sliding, each patch will give you a scalar). Then, you do the same thing for the 2nd kernel though. Now you have two 3x3 feature maps and you stack those together. So, the result of using the 2 kernels on the 5x5x3 image is a 3x3x2 feature map.
    Hey, guys, i met a problem, when i use sess.run(), i found my validation set runs very more slowly than training set. i didn't use optimizer on the validation set, i want to know what happened. Another problem is that, the time i run sess.run([aaa,bbb,ccc]) approximately equal to the time run sess.run(aaa) or bbb or ccc, sess.run() operation is in parallel?
    Hi guys, anyone know how to get reproducible results in tensorflow?
    Akshay Mankar
    @oiwio yes tensorflow framework by default uses parallel computing
    I have a problem when using bazel to 'bazel run' the model in models zoom, i.e., github.com/models/domain_adaptation/. Though I follow the step one by one, it outputs error infomation like, 'ERROR: D:/workspace/slim/BUILD:56:12: in deps attribute of py_library rule //slim:download_and_convert_cifar10: file '//t ensorflow:tensorflow' is misplaced here (expected no files).' I am using Windows 10 x64 + Bazel 0.54.
    Shiva Manne
    Hi guys,
    I have been working on benchmarking commonly used frameworks/libraries for unsupervised learning of word embeddings(word2vec). Since learning embeddings is a frequently used technique, this will be helpful for many working in this field.
    I am currently comparing tensorflow(cpu/gpu), gensim, deeplearning4j and the original c code on standard metrics like training time, peak memory usage and quality of learned vectors.
    Link to my github repo(still working on it).
    I have directly picked up the code for training on each framework from the example given in their respective official github repositories. I ran the benchmark on text8 corpus(plan to run it on a much larger corpus later for the true picture) which gave me strange results.
    I would really appreciate it if you could have a look at the tensorflow code (for word2vec) and give feedback/suggest changes.
    Thanks for your time!
    What is the purpose of combining Max pooling with 1x1 convolution ? http://iamaaditya.github.io/2016/03/one-by-one-convolution/
    anyone familiar with magenta here? I run it on mac with docker and trying to generate from my own dataset
    I successfully built the dataset but when I start the training , it seems to start with logs like INFO:tensorflow:Starting training loop...
    INFO:tensorflow:Create CheckpointSaverHook. but remains stuck for hours
    my load avg is 0.95 on the docker instance so i dont think it s a cpu issue
    Hi.. Do you know any good playground for RNN ? Like this http://playground.tensorflow.org/
    Vaibhav Satam
    Hi guys, I am using estimator api in tensroflow. In input function I am reading csv using decode_csv function and its working perfectly but one thing which I don’t understand is how to do preprocessing of data with this like imputing or trasforming data. I hv seen implementation where ppl are loading data using pandas preprocessing them and feeding them in input function. Whats the best practice for preprocessing data in tensorflow?
    Vaibhav Satam

    I am writing to ask the principle of how to feed a big training data to a tensor flow model. My training data is hosted in csv file(s), and basically I am using the below code to load data in queue.

    filename_queue = tf.train.string_inputproducer([...])
    reader = tf.TextLineReader()
    , line = reader.read(filename_queue)

    line = tf.decode_csv(line, record_defaults=default)
    label_batch, feature_batch = tf.train.shuffle_batch([label, feature], batch_size=batch_size, capacity=512, min_after_dequeue=256, num_threads=8)

    Does anyone have ideas about fixing this: tensorflow/tensorflow#12522
    Trevor R.H. Clarke
    trying to use TFslim to evaluate an inception_v4 net
    I'm classifying a number of images and need to evaluate repeatedly but I won't have all the images at once so I can't create a single batch and call evaluate_once
    I can repeatedly call evaluate_once which works but reloads the checkpoint and reconfigures the net each time which is slow
    can someone point me to a way to load the checkpoint once then set the input batch differently and eval the net each time using tfslim? or do I need to use raw tensorflow to do this?
    Saurabh Vyas
    can anyone, please help me with tensorflow datasetapi ? I am wanting to create a simple dataset for speech recognition, each component consists of mfcc , and target transcription , there is one problem , mfcc is not implemented by default in tensorflow, so I am using python implementation using tf.py_func, but I am getting a strange error
    UnimplementedError: Unsupported object type Tensor
         [[Node: PyFunc = PyFunc[Tin=[DT_STRING, DT_STRING], Tout=[DT_DOUBLE, DT_STRING], token="pyfunc_7"](arg0, arg1)]]
         [[Node: IteratorGetNext_7 = IteratorGetNext[output_shapes=[<unknown>, <unknown>], output_types=[DT_DOUBLE, DT_STRING], _device="/job:localhost/replica:0/task:0/cpu:0"](OneShotIterator_7)]]
    A friend have some project code dev. in TFlow (salary based ),
    if anyone interested, please pm me directly. thanks !
    Caused by: java.io.InvalidClassException: org.apache.spark.unsafe.types.UTF8String; local class incompatible: stream classdesc serialVersionUID = -2992553500466442037, local class serialVersionUID = -5670082246090726217
    Hello when i run spark-shell
    I came across this problem
    Some says it is because the hadoop version of spark .
    However i don't know how to specify the version of hadoop when compile spark source code! any suggestion would be appreciated!
    Loreto Parisi
    anyone is aware of ONNX open model support in TF?
    some import/export has done, but for TF nothing official yet, https://github.com/onnx/tutorials
    please help me on my dummy question: what is the difference between tf.maximum and tf.reduce_max? Are their derivatives different from each other? If I implement maxout/minout, should I use tf.maximum or tf.reduce_max?
    Jay Kim (Data Scientist)
    Hi everyone. I joined this room first time today, nice to meet you all
    Jay Kim (Data Scientist)
    anyone knows how to run tensorflow-on-spark ?