aayux

@dennybritz: I have only tested it on a small subset of the data. The program itself runs smoothly. However, I have not been able to do any meaningful performance/accuracy evaluations...

@chiragnagpal yes, weight decay is largely ineffective when used with Adam. I'm only suggesting a modification on what is currently implemented. We could possibly rethink the entire implementation as proposed...

Sure. It's only something to think about.

The filter width in the convolution layer(s) is effectively acting as an n-gram window. Therefore, by setting `filter_sizes` to `[3, 4, 5]` the kernel is looking at 3-, 4- and...

`tf.nn.embedding_lookup` is an implementation of `tf.gather` which was not supported on GPU at the time of the blog post. Following [this](https://github.com/tensorflow/tensorflow/issues/2502) issue, `tf.gather` works on GPU. So similar changes must...

To get some speedup, set `trainable=False` in `self.W = tf.Variable(...)`. You don't need to train pre-trained embeddings.

Is it possible for you to share the traceback for this error?

> > when you run that command plz try this one > > python2 convfeatures.py --data_path Dataset/flickr30k-images/ --inception_path ConvNets/inception_v4.pb > > i hope this will work fine. > > thank...

@jishan900: Any discussions in this thread should be limited to the title, and this is not a related issue. We'd be happy to help but you must please open a...

Hi, Yes, this was written in Python 2.7 and has only been tested for TensorFlow r1.0 Unfortunately we did not freeze a requirements file so it's really hard to give...