keystone

keystone copied to clipboard

Move SparkContext -> SparkSession

2

We are using SparkContext throughout loaders and example pipelines. It makes sense to move these to using SparkSession given that we're relying on Spark 2.0.

etrain

Lemmatization

4

May be unnecessary for Release 0.1

etrain

Convolution Tradeoffs

1

tomerk

The Least Squares Estimator should have the option to use a local least squares solver. A cost model needs to be built for it, and it needs to be added...

tomerk

Add unit test for block weighted solver with unequal class sizes

1

Relevant discussion on #175

etrain

Create a union operator

1

In addition to the gather operator which concatenates horizontally, it would be nice to have a union operator which concatenates veritcally. This came up in a discussion with @shivaram today...

etrain

Currently transformers are maps from `A => B`, but if I have some complex type like `(Int, Image)` I can't pass it into an `Image => DenseVector[Double]` Transformer and expect...

etrain

tomerk

WeightedBlockSolver for multiple lambdas

1

We often perform a sweep over lambdas to figure out the right regularization -- it matters a lot. @shivaram mentioned that it's possible for the weighted solver to potentially solve...

ericmjonas

Add faster Image Interface - use it to optimize Pooling, Convolution

3

We don't necessarily iterate over images in a couple of common cases in a fast way, e.g. ``` scala while (x < numRows) { while (y < numCols) { while(c...

etrain