Utils

Module with general utilities and missing standard library features that could be useful in any Julia project

Retro-fitted super type of all bit vectors

Custom CUDA version of BitVector (lacking lots of functionality, just a container for now).

Literals are represented as 32-bit signed integers. Positive literals are positive integers identical to their variable. Negative literals are their negations. Integer 0 should not be used to represent literals.

Variables are represented as 32-bit unsigned integers

Find a type that can capture all column values

Are the given sets disjoint (no shared elements)?

Marginalize out dimensions dims from log-probability tensor

An array of 100% probabilities for the given element type

Returns an array of DataFrames where each DataFrame is randomly sampled from the original dataset data

Create mini-batches

Batch size of the dataset

Normalize the given log-likelihood as bits per pixel in data

Retrieve chunks of bit vector

Iterate over columns, excluding the sample weight column

Get the ith example

Get the ith feature values

Computer the per-example log-likelihood of a fully factorized ML model on Bool data

Retrieve the jth bit from a BitVector chunk

Get the weights from a weighted dataset.

Return a copy of Imputed values of X (potentially statistics from another DataFrame)

For example, to impute using same DataFrame:

impute(X; method=:median)

If you want to use another DataFrame to provide imputation statistics:

impute(test_x, train_x; method=:mean)

Supported methods are :median, :mean, :one, :zero

An array of undetermined values (fast) for the given element type

Is the dataset batched?

Is the dataset binary?

Is the data complete (no missing values)?

Is the data column complete (no missing values)?

Is the dataset consisting of floating point data?

Check whether data resides on the GPU

Is the argument not nothing?

Is the dataset weighted?

Convert a literal its variable, removing the sign of the literal

Normalize the given log-likelihood by the number of examples in data

make_missing_mcar(d::DataFrame; keep_prob::Float64=0.8)

Returns a copy of dataframe with making some features missing as MCAR, with keep_prob as probability of keeping each feature.

Compute the marginal prob of each feature in a binary dataset.

An array of 0% probabilities for the given element type

Function that does nothing

For binary complete data, how many UInt64 bit strings are needed to store one feature?

num_examples(df::DataFrame)

Number of examples in data

num_features(df::DataFrame)

Number of features in the data

Number of variables in the data structure

Order the arguments in a tuple in ascending order

Push element into random position in vectorv

Randomly draw samples from the dataset with replacement

Ensure that x resides on the same device as data

shuffle_examples(df::DataFrame)

Shuffle the examples in the data

Reuse a given array if it has the right type and size, otherwise make a new one

Turn binary data into floating point data close to 0 and 1.

Split a weighted dataset into unweighted dataset and its corresponding weights.

Replacement for BitSet.⊆ that does not allocate a new BitSet

Threshold a numeric dataset making it binary

Move data to the CPU

Move data to the GPU

An array of uniform probabilities

Convert a variable to the corresponding positive literal

Get the BitSet of variables in the data structure

Create a weighted copy of the data set