Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
Repo info
    Shikhar Srivastava
    Any help would be appreciated. : )
    Ilari Pihlajisto

    @soilad You can use tf.train.write_graph to save the graph but that doesn't save the weights. Tensorflow has a freeze_graph tool to combine the weights with the graph def: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/freeze_graph.py

    Alternatively you can assign the weights as const values to trainable variables before exporting the graph:

    # Add a new "assign_trained_variables" operation which fills the
    # weights and biases with constant values (trained values)
    ops = []
    for v in tflearn.variables.get_all_trainable_variable():
        vc = tf.constant(v.eval(session=model.session))
        ops.append(tf.assign(v, vc))
    # When using the exported graph, the "assign_trained_variables" op must be run first
    tf.group(*ops, name="assign_trained_variables")
    Shikhar Srivastava
    Thank you @ilaripih . I'll try these out.
    @ilaripih I am struggling for duplicating a network when some condition is met (e.g., performance > threshold). This way, even if the training gets worse, I can recover an intermediate snapshot of the network. Please, any idea how to achieve this?
    Aymeric Damien
    DNN class accept argument 'best_checkpoint_path' and 'best_val_accuract' to snapshot only the best models (http://tflearn.org/models/dnn/#deep-neural-network-model)
    Oliver Gindele

    Hey, I have a question about feed_dics in tflearn. In tensorflow the feed_dict will load the data at each step, so I can preprocess data in batches. Tflearn, however loads the data only once into the feed_dict. Is this correct? I have a function that prepossess the data for each batch. when I use it as input in a tflearn.trainer it only preprocesses the first batch.

    trainer.fit(feed_dicts=self.feed_dict(), val_feed_dicts=self.val_feed_dict(),
                        n_epoch=40, show_metric=True)

    where the feed_dict() function returns a feed dict for each batch. Any ideas how to solve this in tflearn?

    Neil Seward
    Is there a way to batch predictions? My evaluation set is too large and causes memory issues. But I can run the predictions with a smaller subset no problem.
    I am developing a batch evaluation function now, but having something already set up would be great.
    Could anyone help me out for this not-found-tensor-name-in-checkpoint-file issue tflearn/tflearn#527 ? Thanks a lot!
    K Kishore
    When using DNNClassifier, based on what information, we should create hidden_units ?

    I want Use TFRecords for celeba face dataset. celeba face datase has 5 landmark locations, 40 binary attributes annotations per image.

    TFRecords only can give one lable. but I want save all this labels in TFRecords. How can do this in tensorflow?

    Carl Thomé
    How do I use batch_normalization() with conv_3d()? Does it default to only doing mean/var normalization over the number of filters?
    Abdelrahman Elogeel
    Hi, wanted to get idea on how TFLearn would be different than TF-Slim over the long term? My understanding that both provide higher level API for TensorFlow. Thanks!
    Youssef Barhomi
    hello, I am looking for alexnet weights that I can use with tflearn and all I can find is this: http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/ but it seems like half the weights are missing starting at conv2 (conv2 shape is (5, 5, 48, 256) instead of (5, 5, 96, 256)). Does it have to do with how the first alexnet was implemented (using 2 gpus starting at conv2)? thanks!
    @barhomi Yes, it is implemented as original Alexnet (2 froups at conv2)
    you need to call Alexnet(init_model='path/to/weights.np', num_classes=1000, nul_layers_to_init=8)
    Reinis Veips
    I'm trying a hyperparameter optimization with the following code: https://gitlab.com/snippets/1548245
    however, I'm running out of GPU memory (on 12GB K80) at 10th iteration
    it seems to be related to tensorflow/tensorflow#1727
    does anyone know a solution/workaround for this issue?
    Furiously Curious
    For those with multiple NVidia cards: Does adding a SLI bridge help with training time in any way? In other words, does cuDNN detect and use SLI bridge for signaling between multiple cards?
    Aymeric Damien
    @FuriouslyCurious SLI bridge doesn't help. If you want to use multiple GPU, you need to set it up by yourself (data parallelization for example).
    I am using TFLearn and tensorflow 1.0, whenever I try to save a model, I get the following error: WARNING:tensorflow:Error encountered when serializing layer_tensor/LSTM.
    Type is unsupported, or the types of the items don't match field type in CollectionDef.
    'list' object has no attribute 'name'
    Anyone know what I need to resolve this issue?
    Jongyeol Yang
    Hey guys, I have implemented LSTM for simple mnist data using tflearn. I hope it can helpful to understand how to implement LSTM in tflearn.
    Thank you. Very helpful. Does anyone have an example of saving a trained model in tflearn in python and trying to load the trained model in C++ for prediction?
    I saw earlier discussion loading using C++ earlier, but wondering if there's an end to end example using the latest tflearn and tensorflow release.
    Abolfazl Madani
    hi dear all, im coding with python, but i have an error when i def function, and when its running i have this error:
    "Traceback most recent call last"
    please help me
    Andrew H. Johnston
    @abolfazl_madani_twitter Dude that's not the part of the error people need to help you
    That's how all python errors begin
    is there a way to add images in tensorboard through the tflearn api?
    Hi dear all.is there a good way to get the inverse data in tensorflow ? For example,there is a number a,I want to get the inverse type.I found this function in numpy,it is "bumpy.linalg.inv",but this function can only manipulate the data as type of array.when I tried to manipulate the tensor type,error happens
    Ricardo Bartolomeu
    Hi everyone is there a way to remove noise from images using tflearn, (I'm a bit a of a new comer to deep learning)
    For example
    What are some best practices for turning a string value into a number in order to be used by TF? In my example the feature is a domain name and can be one of several million values. I am considering an index in redis and also possibly a hashed value. Thanks.
    Aymeric Damien
    @r3db For noise remover, you can have a check at the auto-encoder examples.
    Pretty new to tflearn i've tried very hard to find a tutorial on object detection(drawing bounding boxes). I can't seem to find how to implement this. I can only find explanations of how to technology works. I've seen many pre-made networks that have this capability but I have a lack of understanding when it comes to figuring out how they implemented it. Can anyone point me in the right direction.?
    Jules Gagnon-Marchand
    Hey guys, I'm starting a "tensorflow-help" channel at https://gitter.im/tensorflow-help/Lobby . Just wanted to do some non-invasive friendly advertisement, once
    Jared Jolton
    Hello! Anyone around? I'm struggling to format time series data for use with tflearn. Wishfully hoping someone in here is an expert :)
    @FreakTheMighty , hello, have you got the good result from the fcn_vgg? I mean the same output size?
    @FreakTheMighty @barhomi hi guys, i have fixed the problem in the fcn_example of Jesse's. Jesse just made a mistake when he did the prediction in the layer, and the upscore layer had no problems. And the output is bad which is quite reasonable, because the weight was trained from dataset which we even don't know, we need to use our dateset to train in for the finetuning. By the way, i just fixed the problem in Jesse test part, and i will rewrite the next part to match the interface with the tflearn.
    Sandip Gangakhedkar
    Hi..has anyone managed to get TFLearn installed on a Ubuntu 16.04 system with CuDNN v6 and Tensorflow 1.4.1?
    I have single directory, Dataset,which contains sub-folders(labels/classes) of images.
    I want to split the Dataset into train and test set for model.fit_generotar().
    How to do that?
    Aaditya Ura

    Hi, I wanted to know here is one tutorial about Classification , But its working on numerical data , Now i f i have a csv with column 0 as categorial data and other columns are having keywords , I want to classify sentence basis on those keywords , I know i can do this with text classifier but is there any way i can do text classification with using same logic of this tutorial ?


    Hello, guys. Simple and possibly repetitive question: I do pip install git+https://github.com/tflearn/tflearn.git, what version of tf it expects? I see 1.1 on the website, some words are there about that I can use the latest tensorflow too, so I use 1.7. But if I try to run some code it shows errors that look like API incompatibility issues, like if tflearn wants some thing older then 1.7.
    Umar Sohail
    Is there anyway to buuld tensorflow-gpu r1.14 with cuda 8 ?
    • What is eager tensor?
    • tf.placeholder is deprecated? what replaced it?
    • How to find whether a given item is a tensor, variable or constant?
    • Some methods in variables are duplicated with prefix R; what does it mean?
    • Many methods are related to scatter what does it (scatter methods) mean?
    • what is diff. between tensor and variable? how to define tensor? never seen tf.Tensor(xyz), while tf.constant/variable(xyz) is quite common