profile
viewpoint
If you are wondering where the data of this site comes from, please visit https://api.github.com/users/ngc92/events. GitMemory does not store any data, but only uses NGINX to cache data for a period of time. The idea behind GitMemory is simply to give users a better reading experience.

danielknobe/blobbyvolley2 68

Official continuation of the famous Blobby Volley 1.x arcade game.

ngc92/quadgym 7

OpenAI gym Environments for Quadrotor Control

ngc92/DeepQLearn 1

deep q learning with experience replay

ngc92/Generative-Adversarial-Networks 1

Implementation of several Generative Adversarial Networks in tensorflow.

ngc92/gym 1

A toolkit for developing and comparing reinforcement learning algorithms.

Mesocopic/branchedflowsim 0

A software package for the simulation of branched flows on random media.

ngc92/baselines 0

OpenAI Baselines: high-quality implementations of reinforcement learning algorithms

ngc92/Box2D 0

Box2D is a 2D physics engine for games

ngc92/branchedflowsim 0

A software package for the simulation of branched flows on random media.

PR opened tzutalin/labelImg

adds a utility function to hide a QT5/QT4 discrepancy

The old code needed to trim the returned text at multiple places. This method changed names between QT4 and QT5, so I've consolidated this into a utility function. The implementation of this function is chosen at import time.

+25 -19

0 comment

2 changed files

pr created time in 2 months

push eventngc92/labelImg

erik

commit sha 39b2ad6591672a88f99a329eef7149740a6aa3d8

adds a utility function to hide a QT5/QT4 discrepancy

view details

push time in 2 months

fork ngc92/labelImg

🖍️ LabelImg is a graphical image annotation tool and label object bounding boxes in images

https://youtu.be/p0nR2YsCY_U

fork in 2 months

issue commenttensorflow/tensorflow

Partial shapes for keras.InputLayer with sparse=True

It appears that this works now. It is also shown in the documentation here https://www.tensorflow.org/guide/sparse_tensor#tfkeras

ngc92

comment created time in 2 months

issue closedtensorflow/tensorflow

Partial shapes for keras.InputLayer with sparse=True

System information

  • TensorFlow version (you are using): 2.1
  • Are you willing to contribute it (Yes/No): Yes

Describe the feature and the current behavior/state. It is currently not possible to use input SparseTensor into e.g. keras.Sequential with partially defined shapes. This is because SparseTensors need to be declared specifically, thus necessitating the use of keras.InputLayer as the fiirst layer in the model.

If the shape of keras.InputLayer for sparse inputs is not fully defined (including the batch size), the produced SparseTensor will have dense_shape where only the rank is defined. This happens because unless the dense_shapeis fully defined, a placeholder will be used for feeding in the dense_shape, removing the partial shape information.

Will this change the current api? How? No

Who will benefit with this feature? This will make models of the form

Sequenctial([
   InputLayer(shape=(num_features,), sparse=True), 
  Dense(num_units)
])

work, as they require partial shape for the input of the Dense layer.

closed time in 2 months

ngc92