gradientinstitute / aboleth

A bare-bones TensorFlow framework for Bayesian deep learning and Gaussian process approximation

gradientinstitute/aboleth Issues

NCP needs a "pass-through" dimension argument
Closed 6 years ago
Consolidate the imputation layers
Closed 6 years ago
Low-rank style imputation
Updated 6 years ago
Implement Flipout sampling?
Updated 6 years ago
Implement Noise Contrastive Priors for better uncertainty estimates
Closed 6 years ago
What is the dtype in the Dense Layers?
Updated 6 years ago
Investigate using custom estimators
Closed 6 years ago
Allow independent samples in the DropOut layer
Closed 6 years ago
Implement and investigate a shared stddev posterior on weights for variational layers
Closed 6 years ago1
Improve imputation
Closed 6 years ago
What function should we be using inside "util.pos"
Closed 6 years ago
Low-rank, joint imputation layers
Closed 6 years ago2
Layer shape compatibility check
Updated 6 years ago1
Replace distributions.py with TensorFlow probability trainable_distributions?
Closed 6 years ago1
Remove MAP layers where possible in favour of using Keras layers?
Closed 6 years ago1
Variable MC samples specified with placeholders causes issues for Reshape layer
Closed 6 years ago2
Initialisation shapes are inconsistent
Closed 6 years ago1
Implement "autonorm" for MAP layers, and auto-norm dropout
Closed 6 years ago1
Initialize MAP weights with glorat_uniform
Closed 6 years ago
Remove interface for incremental updates
Closed 6 years ago
Change interface for variational layers
Closed 6 years ago
Make a VAE?
Updated 6 years ago2
Refactor imputation module
Closed 6 years ago
Can n_samples be a placeholder?
Closed 7 years ago
Make it more obvious in the variational layers what is being optimised
Closed 6 years ago1
Allow batching of model draws during predictions
Closed 7 years ago
Make it clear python 2 is not supported
Closed 6 years ago
Make sure all variables are view-able through tensorboard
Closed 6 years ago
Turn prior variable optimisation *off* by default?
Closed 6 years ago
Add a categorical imputer
Closed 6 years ago
Pass in likelihood as a tensor to objectives, rather than an object
Closed 6 years ago
Generalized feature embedding
Closed 7 years ago
Apply a bit more DRY in layers.py
Closed 7 years ago
Checkpoint best model instead of n most recent
Closed 7 years ago4
Tutorial on building regressors with Aboleth
Closed 7 years ago1
Test values output from layers in the unit tests
Updated 7 years ago
Use more placeholders in tests
Updated 7 years ago
Check shape of training weights to ensure correct broadcast
Closed 7 years ago2
Make a predictions convenience module
Closed 7 years ago
Make Conv2D MAP layer
Closed 7 years ago1
Migrate Demos Datasets to Tensorflow 1.4
Closed 7 years ago1
Make impute layers use input dtype and not default to float32
Closed 7 years ago
Why does the loss suddenly increase when the queue runners terminate?
Closed 7 years ago1
Use Python 3 Type hinting and mypy static type checking
Updated 7 years ago1
Further investigate the log determinant calculation in distributions.py
Closed 7 years ago
Tutorial on embedding with Aboleth
Updated 7 years ago
Tutorial on imputation with Aboleth
Updated 7 years ago
Things to clarify in Aboleth
Closed 7 years ago
Investigate tf.distributions to replace our distributions in distributions.py
Closed 7 years ago
Look at the TensorFlow DatasetAPI to replace the queues in the Demos
Closed 7 years ago