Merge pull request #22 from JuliaAI/dev

For a 0.3.0 release

Author: ablaom

Project.toml

@@ -1,7 +1,7 @@
 name = "LearnTestAPI"
 uuid = "3111ed91-c4f2-40e7-bb19-7f6c618409b8"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.2.4"
+version = "0.3.0"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
@@ -25,13 +25,13 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 UnPack = "3a884ed6-31ef-47d7-9d2a-63182c4928ed"
 
 [compat]
-CategoricalArrays = "0.10.8"
-CategoricalDistributions = "0.1.15"
+CategoricalArrays = "1"
+CategoricalDistributions = "0.2"
 Distributions = "0.25"
 InteractiveUtils = "<0.0.1, 1"
 IsURL = "0.2.0"
-LearnAPI = "0.2.0,1"
-LearnDataFrontEnds = "0.1"
+LearnAPI = "2"
+LearnDataFrontEnds = "0.2"
 LinearAlgebra = "<0.0.1, 1"
 MLCore = "1.0.0"
 MacroTools = "0.5"

src/learners/classification.jl

@@ -24,7 +24,7 @@ struct ConstantClassifierFitted
     learner::ConstantClassifier
     probabilities
     names::Vector{Symbol}
-    classes_seen
+    levels_seen
     codes_seen
     decoder
 end
@@ -44,10 +44,15 @@ LearnAPI.features(learner::ConstantClassifier, data) =
 LearnAPI.target(learner::ConstantClassifier, data) =
     LearnAPI.target(learner, data, front_end)
 
-function LearnAPI.fit(learner::ConstantClassifier, observations::FrontEnds.Obs; verbosity=1)
+function LearnAPI.fit(
+    learner::ConstantClassifier,
+    observations::FrontEnds.Obs;
+    verbosity=LearnAPI.default_verbosity(),
+    )
+
     y = observations.target # integer "codes"
     names = observations.names
-    classes_seen = observations.classes_seen
+    levels_seen = observations.levels_seen
     codes_seen = sort(unique(y))
     decoder = observations.decoder
 
@@ -59,7 +64,7 @@ function LearnAPI.fit(learner::ConstantClassifier, observations::FrontEnds.Obs;
         learner,
         probabilities,
         names,
-        classes_seen,
+        levels_seen,
         codes_seen,
         decoder,
     )
@@ -89,7 +94,7 @@ function LearnAPI.predict(
     probs = model.probabilities
     # repeat vertically to get rows of a matrix:
     probs_matrix = reshape(repeat(probs, n), (length(probs), n))'
-    return CategoricalDistributions.UnivariateFinite(model.classes_seen, probs_matrix)
+    return CategoricalDistributions.UnivariateFinite(model.levels_seen, probs_matrix)
 end
 LearnAPI.predict(model::ConstantClassifierFitted, ::Distribution, data) =
         predict(model, Distribution(), obs(model, data))

src/learners/dimension_reduction.jl

@@ -65,7 +65,11 @@ LearnAPI.obs(model::TruncatedSVDFitted, data) =
 LearnAPI.features(learner::TruncatedSVD, data) =
     LearnAPI.features(learner, data, FrontEnds.Tarragon())
 
-function LearnAPI.fit(learner::TruncatedSVD, observations::FrontEnds.Obs; verbosity=1)
+function LearnAPI.fit(
+    learner::TruncatedSVD,
+    observations::FrontEnds.Obs;
+    verbosity=LearnAPI.default_verbosity(),
+    )
 
     # unpack hyperparameters:
     codim = learner.codim

src/learners/ensembling.jl

@@ -69,7 +69,7 @@ LearnAPI.obs(model::EnsembleFitted, data) = LearnAPI.obs(first(model.models), da
 LearnAPI.target(learner::Ensemble, data) = LearnAPI.target(learner.atom, data)
 LearnAPI.features(learner::Ensemble, data) = LearnAPI.features(learner.atom, data)
 
-function LearnAPI.fit(learner::Ensemble, data; verbosity=1)
+function LearnAPI.fit(learner::Ensemble, data; verbosity=LearnAPI.default_verbosity())
 
     # unpack hyperparameters:
     atom = learner.atom
@@ -112,7 +112,7 @@ function LearnAPI.update(
     model::EnsembleFitted,
     data,
     replacements::Pair{Symbol}...;
-    verbosity=1,
+    verbosity=LearnAPI.default_verbosity(),
     )
     learner_old = LearnAPI.learner(model)
     learner = LearnAPI.clone(learner_old, replacements...)
@@ -205,7 +205,7 @@ LearnAPI.components(model::EnsembleFitted) = [:atom => model.models,]
 # - `update`
 # - `predict` (`Point` predictions)
 # - `predictions` (returns predictions on all supplied data)
-# - `out_of_sample_indices` (articluates which data is the internal validation data)
+# - `out_of_sample_indices` (articulates which data is the internal validation data)
 # - `trees`
 # - `training_losses`
 # - `out_of_sample_losses`
@@ -361,7 +361,7 @@ struct StumpRegressorFitted
     rng
 end
 
-function LearnAPI.fit(learner::StumpRegressor, data; verbosity=1)
+function LearnAPI.fit(learner::StumpRegressor, data; verbosity=LearnAPI.default_verbosity())
 
     x, y = data
     rng = deepcopy(learner.rng)
@@ -426,7 +426,7 @@ function LearnAPI.update(
     model::StumpRegressorFitted,
     data, # ignored as cached
     replacements::Pair{Symbol}...;
-    verbosity=1,
+    verbosity=LearnAPI.default_verbosity(),
     )
 
     learner_old = LearnAPI.learner(model)
@@ -490,8 +490,9 @@ function LearnAPI.update(
 
 end
 
-# needed, because model is supervised:
-LearnAPI.target(learner::StumpRegressor, observations) = last(observations)
+# training data deconstructors:
+LearnAPI.features(learner::StumpRegressor, data) = first(data)
+LearnAPI.target(learner::StumpRegressor, data) = last(data)
 
 LearnAPI.predict(model::StumpRegressorFitted, ::Point, x) =
     _predict(model.forest, x)

src/learners/gradient_descent.jl

@@ -10,6 +10,7 @@ using StableRNGs
 import Optimisers
 import Zygote
 import NNlib
+import CategoricalArrays
 import CategoricalDistributions
 import CategoricalDistributions: pdf, mode
 import ComponentArrays
@@ -55,7 +56,7 @@ for the specified number of `epochs`.
 - `perceptron`: component array with components `weights` and `bias`
 - `optimiser`: optimiser from Optimiser.jl
 - `X`: feature matrix, of size `(p, n)`
-- `y_hot`: one-hot encoded target, of size `(nclasses, n)`
+- `y_hot`: one-hot encoded target, of size `(nlevels, n)`
 - `epochs`: number of epochs
 - `state`: optimiser state
 
@@ -108,7 +109,7 @@ point predictions with `predict(model, Point(), Xnew)`.
 
 # Warm restart options
 
-    update(model, newdata, :epochs=>n, other_replacements...; verbosity=1)
+    update(model, newdata, :epochs=>n, other_replacements...)
 
 If `Δepochs = n - perceptron.epochs` is non-negative, then return an updated model, with
 the weights and bias of the previously learned perceptron used as the starting state in
@@ -117,7 +118,7 @@ instead of the previous training data. Any other hyperparaameter `replacements`
 adopted. If `Δepochs` is negative or not specified, instead return `fit(learner,
 newdata)`, where `learner=LearnAPI.clone(learner; epochs=n, replacements....)`.
 
-    update_observations(model, newdata, replacements...; verbosity=1)
+    update_observations(model, newdata, replacements...)
 
 Return an updated model, with the weights and bias of the previously learned perceptron
 used as the starting state in new gradient descent updates. Adopt any specified
@@ -132,38 +133,38 @@ PerceptronClassifier(; epochs=50, optimiser=Optimisers.Adam(), rng=Random.defaul
 struct PerceptronClassifierObs
     X::Matrix{Float32}
     y_hot::BitMatrix  # one-hot encoded target
-    classes           # the (ordered) pool of `y`, as `CategoricalValue`s
+    levels           # the (ordered) pool of `y`, as `CategoricalValue`s
 end
 
 # For pre-processing the training data:
 function LearnAPI.obs(::PerceptronClassifier, data::Tuple)
     X, y = data
-    classes = CategoricalDistributions.classes(y)
-    y_hot = classes .== permutedims(y) # one-hot encoding
-    return PerceptronClassifierObs(X, y_hot, classes)
+    levels = CategoricalArrays.levels(y)
+    y_hot = levels .== permutedims(y) # one-hot encoding
+    return PerceptronClassifierObs(X, y_hot, levels)
 end
 LearnAPI.obs(::PerceptronClassifier, observations::PerceptronClassifierObs) =
     observations # involutivity
 
 # helper:
-function decode(y_hot, classes)
+function decode(y_hot, levels)
     n = size(y_hot, 2)
-    [only(classes[y_hot[:,i]]) for i in 1:n]
+    [only(levels[y_hot[:,i]]) for i in 1:n]
 end
 
 # implement `RadomAccess()` interface for output of `obs`:
 Base.length(observations::PerceptronClassifierObs) = size(observations.y_hot, 2)
 Base.getindex(observations::PerceptronClassifierObs, I) = PerceptronClassifierObs(
     observations.X[:, I],
     observations.y_hot[:, I],
-    observations.classes,
+    observations.levels,
 )
 
 # training data deconstructors:
 LearnAPI.target(
     learner::PerceptronClassifier,
     observations::PerceptronClassifierObs,
-) = decode(observations.y_hot, observations.classes)
+) = decode(observations.y_hot, observations.levels)
 LearnAPI.target(learner::PerceptronClassifier, data) =
     LearnAPI.target(learner, obs(learner, data))
 LearnAPI.features(
@@ -184,7 +185,7 @@ struct PerceptronClassifierFitted
     learner::PerceptronClassifier
     perceptron  # component array storing weights and bias
     state       # optimiser state
-    classes     # target classes
+    levels     # target levels
     losses
 end
 
@@ -194,7 +195,7 @@ LearnAPI.learner(model::PerceptronClassifierFitted) = model.learner
 function LearnAPI.fit(
     learner::PerceptronClassifier,
     observations::PerceptronClassifierObs;
-    verbosity=1,
+    verbosity=LearnAPI.default_verbosity(),
     )
 
     # unpack hyperparameters:
@@ -205,20 +206,20 @@ function LearnAPI.fit(
     # unpack data:
     X = observations.X
     y_hot = observations.y_hot
-    classes = observations.classes
-    nclasses = length(classes)
+    levels = observations.levels
+    nlevels = length(levels)
 
     # initialize bias and weights:
-    weights = randn(rng, Float32, nclasses, p)
-    bias = zeros(Float32, nclasses)
+    weights = randn(rng, Float32, nlevels, p)
+    bias = zeros(Float32, nlevels)
     perceptron = (; weights, bias) |> ComponentArrays.ComponentArray
 
     # initialize optimiser:
     state = Optimisers.setup(optimiser, perceptron)
 
     perceptron, state, losses = corefit(perceptron, X, y_hot, epochs, state, verbosity)
 
-    return PerceptronClassifierFitted(learner, perceptron, state, classes, losses)
+    return PerceptronClassifierFitted(learner, perceptron, state, levels, losses)
 end
 
 # `fit` for unprocessed data:
@@ -230,16 +231,16 @@ function LearnAPI.update_observations(
     model::PerceptronClassifierFitted,
     observations_new::PerceptronClassifierObs,
     replacements...;
-    verbosity=1,
+    verbosity=LearnAPI.default_verbosity(),
     )
 
     # unpack data:
     X = observations_new.X
     y_hot = observations_new.y_hot
-    classes = observations_new.classes
-    nclasses = length(classes)
+    levels = observations_new.levels
+    nlevels = length(levels)
 
-    classes == model.classes || error("New training target has incompatible classes.")
+    levels == model.levels || error("New training target has incompatible levels.")
 
     learner_old = LearnAPI.learner(model)
     learner = LearnAPI.clone(learner_old, replacements...)
@@ -252,7 +253,7 @@ function LearnAPI.update_observations(
     perceptron, state, losses_new = corefit(perceptron, X, y_hot, epochs, state, verbosity)
     losses = vcat(losses, losses_new)
 
-    return PerceptronClassifierFitted(learner, perceptron, state, classes, losses)
+    return PerceptronClassifierFitted(learner, perceptron, state, levels, losses)
 end
 LearnAPI.update_observations(model::PerceptronClassifierFitted, data, args...; kwargs...) =
     update_observations(model, obs(LearnAPI.learner(model), data), args...; kwargs...)
@@ -262,16 +263,16 @@ function LearnAPI.update(
     model::PerceptronClassifierFitted,
     observations::PerceptronClassifierObs,
     replacements...;
-    verbosity=1,
+    verbosity=LearnAPI.default_verbosity(),
     )
 
     # unpack data:
     X = observations.X
     y_hot = observations.y_hot
-    classes = observations.classes
-    nclasses = length(classes)
+    levels = observations.levels
+    nlevels = length(levels)
 
-    classes == model.classes || error("New training target has incompatible classes.")
+    levels == model.levels || error("New training target has incompatible levels.")
 
     learner_old = LearnAPI.learner(model)
     learner = LearnAPI.clone(learner_old, replacements...)
@@ -289,7 +290,7 @@ function LearnAPI.update(
         corefit(perceptron, X, y_hot, Δepochs, state, verbosity)
     losses = vcat(losses, losses_new)
 
-    return PerceptronClassifierFitted(learner, perceptron, state, classes, losses)
+    return PerceptronClassifierFitted(learner, perceptron, state, levels, losses)
 end
 LearnAPI.update(model::PerceptronClassifierFitted, data, args...; kwargs...) =
     update(model, obs(LearnAPI.learner(model), data), args...; kwargs...)
@@ -299,9 +300,9 @@ LearnAPI.update(model::PerceptronClassifierFitted, data, args...; kwargs...) =
 
 function LearnAPI.predict(model::PerceptronClassifierFitted, ::Distribution, Xnew)
     perceptron = model.perceptron
-    classes = model.classes
+    levels = model.levels
     probs = perceptron.weights*Xnew .+ perceptron.bias |> NNlib.softmax
-    return CategoricalDistributions.UnivariateFinite(classes, probs')
+    return CategoricalDistributions.UnivariateFinite(levels, probs')
 end
 
 LearnAPI.predict(model::PerceptronClassifierFitted, ::Point, Xnew) =