# Changelog
## v3.0.2
- do not take mutable reference when subsampling with `into_subsampled`.

## v3.0.1
- new method `into_subsampled` which returns a new ecDNA distribution without modifying the original one.

## v3.0.0
Do not consume the `rng` when subsampling and use a mutable reference instead (we have lost reproducibility in subsampling though).

## v2.0.6
Store in the abc results `kmax`.

## v2.0.5
Compute the quantile using `ABCRejection::compute_quantile_to_get_10_cells` for the `k_max` stat.

## v2.0.4
Increase quantile for `kmax_stat` to `0.999`.

## v2.0.3
Serialise field `kmax_stat` in `ABCResult` with `serde`.

## v2.0.2
Rename field `k_max` in `ABCResult` to `kmax_stat`.

## v2.0.1
### Added
New field `k_max` in `ABCResult`: `hdrhistogram` to compute histogram for abc in order to get the mean ecDNA and the quantile for the max ecDNA copy number.

## v2.0.0
### Added
`ABCRejection::run` take bool argument `drop_nminus` which set the `frequency_stat` to `f32::NAN`.
Added new field `dropped_nminus` in the `ABCResult`.

## v1.0.1
### Added
Add struct `ABCResultsFitness` that can be saved and loaded.

## v0.5.4
### Added
Computation of the variance `EcDNADistribution::compute_variance`.

## v0.5.3
### BugFix
Fix the exponential sampling `k * exp` instead of `k (1 - exp)`.

## v0.5.2
Reintroduce exponential sampling.

## v0.5.1
### BugFix
`Poisson` instead of `Exp` in sampling.

## v0.5.0
### BugFix
We were doing it the exponential noise wrong: this is a Poisson point process with `lambda` which varies for each sample and `k` being the number of copies to transcribed.
This generates a Poisson point process with mean `k*lambda` (assuming same `lambda` for each all copies, that is an homogeneous Poisson point process), see [wikipedia](https://en.wikipedia.org/wiki/Poisson_point_process#Poisson_distribution_of_point_counts).

## v0.4.4
Method `EcDNADistribution::sample` is reproducible.

## v0.4.3
Sort by keys (`DNACopy`) while displaying the `EcDNADistribution`.

## v0.4.2
Implement `Display` for the `EcDNADistribution`.

## v0.4.1
### BugFix
`lambda` of the exponential is `f32` not `NonZeroU8`.
## v0.4.0
The exponential strategy for sampling takes a `NonZeroU8` as parameter and perform the following mapping: `k(1 - r)` where `r` is a random number generated from `Exp(NonZeroU8)`.

## v0.3.5
Implement `SamplingStrategy::Exponential(scale)`. Note that this distribution has the same scale parameters for all cells, that is independently of the number of `k` copies present in cells.

## v0.3.4
Implement `SamplingStrategy::Poisson`.
### Bugfix
Ensure that cells without ecDNAs cannot gain any ecDNA by randomly mapping the distribution to a Gaussian or Poisson.

## v0.3.2
Implement `Eq` for `SamplingStrategy`.
## v0.3.1
Allows `sample` to take a sample with a number of cells equal to the ecDNA distribution size, that is to not undersample.

## v0.3.0
Make `undersample` private and `sample` public which samples the ecDNA distribution according to a `SamplingStrategy`.

## v0.2.2
Make `distribution.create_histogram` public and remove `create_histogram_f32`.

## v0.2.0
Remove the rng `ChaCha` and keep it generic with `impl Rng` trait.

## v0.1.3
### Bugfix
Fix the method `EcDNADistribution::drop_cells_with_k_copies`.

## v0.1.2
- New method `EcDNADistribution::drop_cells_with_k_copies`

## v0.1.1
- Use the Kolmogorov-Smirnov implementation found in [`kolmogorov_smirnov`](https://github.com/daithiocrualaoich/kolmogorov_smirnov/blob/cb067e92ec837efbad66e8bbcf85500ad778feb8/src/test.rs#L127) crate.
- Run the ks test considering only cells with ecDNAs.