-
Notifications
You must be signed in to change notification settings - Fork 20
/
models.jl
518 lines (453 loc) · 13.9 KB
/
models.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
abstract type ModelType end
abstract type GradientRegression <: ModelType end
abstract type MLE2P <: ModelType end # 2-parameters max-likelihood
abstract type MSE <: GradientRegression end
abstract type LogLoss <: GradientRegression end
abstract type Poisson <: GradientRegression end
abstract type Gamma <: GradientRegression end
abstract type Tweedie <: GradientRegression end
abstract type MLogLoss <: ModelType end
abstract type GaussianMLE <: MLE2P end
abstract type LogisticMLE <: MLE2P end
abstract type Quantile <: ModelType end
abstract type L1 <: ModelType end
# Converts MSE -> :mse
const _type2loss_dict = Dict(
MSE => :mse,
LogLoss => :logloss,
Poisson => :poisson,
Gamma => :gamma,
Tweedie => :tweedie,
MLogLoss => :mlogloss,
GaussianMLE => :gaussian_mle,
LogisticMLE => :logistic_mle,
Quantile => :quantile,
L1 => :l1,
)
_type2loss(L::Type) = _type2loss_dict[L]
# make a Random Number Generator object
mk_rng(rng::AbstractRNG) = rng
mk_rng(int::Integer) = Random.MersenneTwister(int)
mutable struct EvoTreeRegressor{L<:ModelType} <: MMI.Deterministic
nrounds::Int
L2::Float64
lambda::Float64
gamma::Float64
eta::Float64
max_depth::Int
min_weight::Float64 # real minimum number of observations, different from xgboost (but same for linear)
rowsample::Float64 # subsample
colsample::Float64
nbins::Int
alpha::Float64
monotone_constraints::Any
tree_type::String
rng::Any
end
function EvoTreeRegressor(; kwargs...)
# defaults arguments
args = Dict{Symbol,Any}(
:loss => :mse,
:nrounds => 100,
:L2 => 0.0,
:lambda => 0.0,
:gamma => 0.0, # min gain to split
:eta => 0.1, # learning rate
:max_depth => 6,
:min_weight => 1.0, # minimal weight, different from xgboost (but same for linear)
:rowsample => 1.0,
:colsample => 1.0,
:nbins => 64,
:alpha => 0.5,
:monotone_constraints => Dict{Int,Int}(),
:tree_type => "binary",
:rng => 123,
)
args_override = intersect(keys(args), keys(kwargs))
for arg in args_override
args[arg] = kwargs[arg]
end
args[:rng] = mk_rng(args[:rng])
args[:loss] = Symbol(args[:loss])
if args[:loss] == :mse
L = MSE
elseif args[:loss] == :linear
L = MSE
elseif args[:loss] == :logloss
L = LogLoss
elseif args[:loss] == :logistic
L = LogLoss
elseif args[:loss] == :gamma
L = Gamma
elseif args[:loss] == :tweedie
L = Tweedie
elseif args[:loss] == :l1
L = L1
elseif args[:loss] == :quantile
L = Quantile
else
error(
"Invalid loss: $(args[:loss]). Only [`:mse`, `:logloss`, `:gamma`, `:tweedie`, `:l1`, `:quantile`] are supported by EvoTreeRegressor.",
)
end
check_args(args)
model = EvoTreeRegressor{L}(
args[:nrounds],
args[:L2],
args[:lambda],
args[:gamma],
args[:eta],
args[:max_depth],
args[:min_weight],
args[:rowsample],
args[:colsample],
args[:nbins],
args[:alpha],
args[:monotone_constraints],
args[:tree_type],
args[:rng],
)
return model
end
function EvoTreeRegressor{L}(; kwargs...) where {L}
EvoTreeRegressor(; loss=_type2loss(L), kwargs...)
end
mutable struct EvoTreeCount{L<:ModelType} <: MMI.Probabilistic
nrounds::Int
L2::Float64
lambda::Float64
gamma::Float64
eta::Float64
max_depth::Int
min_weight::Float64 # real minimum number of observations, different from xgboost (but same for linear)
rowsample::Float64 # subsample
colsample::Float64
nbins::Int
alpha::Float64
monotone_constraints::Any
tree_type::String
rng::Any
end
function EvoTreeCount(; kwargs...)
# defaults arguments
args = Dict{Symbol,Any}(
:nrounds => 100,
:L2 => 0.0,
:lambda => 0.0,
:gamma => 0.0, # min gain to split
:eta => 0.1, # learning rate
:max_depth => 6,
:min_weight => 1.0, # minimal weight, different from xgboost (but same for linear)
:rowsample => 1.0,
:colsample => 1.0,
:nbins => 64,
:alpha => 0.5,
:monotone_constraints => Dict{Int,Int}(),
:tree_type => "binary",
:rng => 123,
)
args_override = intersect(keys(args), keys(kwargs))
for arg in args_override
args[arg] = kwargs[arg]
end
args[:rng] = mk_rng(args[:rng])
L = Poisson
check_args(args)
model = EvoTreeCount{L}(
args[:nrounds],
args[:L2],
args[:lambda],
args[:gamma],
args[:eta],
args[:max_depth],
args[:min_weight],
args[:rowsample],
args[:colsample],
args[:nbins],
args[:alpha],
args[:monotone_constraints],
args[:tree_type],
args[:rng],
)
return model
end
function EvoTreeCount{L}(; kwargs...) where {L}
EvoTreeCount(; kwargs...)
end
mutable struct EvoTreeClassifier{L<:ModelType} <: MMI.Probabilistic
nrounds::Int
L2::Float64
lambda::Float64
gamma::Float64
eta::Float64
max_depth::Int
min_weight::Float64 # real minimum number of observations, different from xgboost (but same for linear)
rowsample::Float64 # subsample
colsample::Float64
nbins::Int
alpha::Float64
tree_type::String
rng::Any
end
function EvoTreeClassifier(; kwargs...)
# defaults arguments
args = Dict{Symbol,Any}(
:nrounds => 100,
:L2 => 0.0,
:lambda => 0.0,
:gamma => 0.0, # min gain to split
:eta => 0.1, # learning rate
:max_depth => 6,
:min_weight => 1.0, # minimal weight, different from xgboost (but same for linear)
:rowsample => 1.0,
:colsample => 1.0,
:nbins => 64,
:alpha => 0.5,
:tree_type => "binary",
:rng => 123,
)
args_override = intersect(keys(args), keys(kwargs))
for arg in args_override
args[arg] = kwargs[arg]
end
args[:rng] = mk_rng(args[:rng])
L = MLogLoss
check_args(args)
model = EvoTreeClassifier{L}(
args[:nrounds],
args[:L2],
args[:lambda],
args[:gamma],
args[:eta],
args[:max_depth],
args[:min_weight],
args[:rowsample],
args[:colsample],
args[:nbins],
args[:alpha],
args[:tree_type],
args[:rng],
)
return model
end
function EvoTreeClassifier{L}(; kwargs...) where {L}
EvoTreeClassifier(; kwargs...)
end
mutable struct EvoTreeMLE{L<:ModelType} <: MMI.Probabilistic
nrounds::Int
L2::Float64
lambda::Float64
gamma::Float64
eta::Float64
max_depth::Int
min_weight::Float64 # real minimum number of observations, different from xgboost (but same for linear)
rowsample::Float64 # subsample
colsample::Float64
nbins::Int
alpha::Float64
monotone_constraints::Any
tree_type::String
rng::Any
end
function EvoTreeMLE(; kwargs...)
# defaults arguments
args = Dict{Symbol,Any}(
:loss => :gaussian_mle,
:nrounds => 100,
:L2 => 0.0,
:lambda => 0.0,
:gamma => 0.0, # min gain to split
:eta => 0.1, # learning rate
:max_depth => 6,
:min_weight => 8.0, # minimal weight, different from xgboost (but same for linear)
:rowsample => 1.0,
:colsample => 1.0,
:nbins => 64,
:alpha => 0.5,
:monotone_constraints => Dict{Int,Int}(),
:tree_type => "binary",
:rng => 123,
)
args_override = intersect(keys(args), keys(kwargs))
for arg in args_override
args[arg] = kwargs[arg]
end
args[:rng] = mk_rng(args[:rng])
args[:loss] = Symbol(args[:loss])
if args[:loss] in [:gaussian, :gaussian_mle]
L = GaussianMLE
elseif args[:loss] in [:logistic, :logistic_mle]
L = LogisticMLE
else
error(
"Invalid loss: $(args[:loss]). Only `:gaussian_mle` and `:logistic_mle` are supported by EvoTreeMLE.",
)
end
check_args(args)
model = EvoTreeMLE{L}(
args[:nrounds],
args[:L2],
args[:lambda],
args[:gamma],
args[:eta],
args[:max_depth],
args[:min_weight],
args[:rowsample],
args[:colsample],
args[:nbins],
args[:alpha],
args[:monotone_constraints],
args[:tree_type],
args[:rng],
)
return model
end
function EvoTreeMLE{L}(; kwargs...) where {L}
if L == GaussianMLE
loss = :gaussian_mle
elseif L == LogisticMLE
loss = :logistic_mle
end
EvoTreeMLE(; loss=loss, kwargs...)
end
mutable struct EvoTreeGaussian{L<:ModelType} <: MMI.Probabilistic
nrounds::Int
L2::Float64
lambda::Float64
gamma::Float64
eta::Float64
max_depth::Int
min_weight::Float64 # real minimum number of observations, different from xgboost (but same for linear)
rowsample::Float64 # subsample
colsample::Float64
nbins::Int
alpha::Float64
monotone_constraints::Any
tree_type::String
rng::Any
end
function EvoTreeGaussian(; kwargs...)
# defaults arguments
args = Dict{Symbol,Any}(
:nrounds => 100,
:L2 => 0.0,
:lambda => 0.0,
:gamma => 0.0, # min gain to split
:eta => 0.1, # learning rate
:max_depth => 6,
:min_weight => 8.0, # minimal weight, different from xgboost (but same for linear)
:rowsample => 1.0,
:colsample => 1.0,
:nbins => 64,
:alpha => 0.5,
:monotone_constraints => Dict{Int,Int}(),
:tree_type => "binary",
:rng => 123,
)
args_override = intersect(keys(args), keys(kwargs))
for arg in args_override
args[arg] = kwargs[arg]
end
args[:rng] = mk_rng(args[:rng])
L = GaussianMLE
check_args(args)
model = EvoTreeGaussian{L}(
args[:nrounds],
args[:L2],
args[:lambda],
args[:gamma],
args[:eta],
args[:max_depth],
args[:min_weight],
args[:rowsample],
args[:colsample],
args[:nbins],
args[:alpha],
args[:monotone_constraints],
args[:tree_type],
args[:rng],
)
return model
end
function EvoTreeGaussian{L}(; kwargs...) where {L}
EvoTreeGaussian(; kwargs...)
end
const EvoTypes{L} = Union{
EvoTreeRegressor{L},
EvoTreeCount{L},
EvoTreeClassifier{L},
EvoTreeGaussian{L},
EvoTreeMLE{L},
}
_get_struct_loss(::EvoTypes{L}) where {L} = L
function Base.show(io::IO, config::EvoTypes)
println(io, "$(typeof(config))")
for fname in fieldnames(typeof(config))
println(io, " - $fname: $(getfield(config, fname))")
end
end
"""
check_parameter(::Type{<:T}, value, min_value::Real, max_value::Real, label::Symbol) where {T<:Number}
Check model parameter if it's valid
"""
function check_parameter(::Type{<:T}, value, min_value::Real, max_value::Real, label::Symbol) where {T<:Number}
min_value = max(typemin(T), min_value)
max_value = min(typemax(T), max_value)
try
convert(T, value)
@assert min_value <= value <= max_value
catch
error("Invalid value for parameter `$(string(label))`: $value. `$(string(label))` must be of type $T with value between $min_value and $max_value.")
end
end
"""
check_args(args::Dict{Symbol,Any})
Check model arguments if they are valid
"""
function check_args(args::Dict{Symbol,Any})
# Check integer parameters
check_parameter(Int, args[:nrounds], 0, typemax(Int), :nrounds)
check_parameter(Int, args[:max_depth], 1, typemax(Int), :max_depth)
check_parameter(Int, args[:nbins], 2, 255, :nbins)
# check positive float parameters
check_parameter(Float64, args[:lambda], zero(Float64), typemax(Float64), :lambda)
check_parameter(Float64, args[:gamma], zero(Float64), typemax(Float64), :gamma)
check_parameter(Float64, args[:min_weight], zero(Float64), typemax(Float64), :min_weight)
# check bounded parameters
check_parameter(Float64, args[:alpha], zero(Float64), one(Float64), :alpha)
check_parameter(Float64, args[:rowsample], eps(Float64), one(Float64), :rowsample)
check_parameter(Float64, args[:colsample], eps(Float64), one(Float64), :colsample)
check_parameter(Float64, args[:eta], zero(Float64), typemax(Float64), :eta)
try
tree_type = string(args[:tree_type])
@assert tree_type ∈ ["binary", "oblivious"]
catch
error("Invalid input for `tree_type` parameter: `$(args[:tree_type])`. Must be of one of `binary` or `oblivious`")
end
end
"""
check_args(model::EvoTypes{L}) where {L}
Check model arguments if they are valid (eg, after mutation when tuning hyperparams)
Note: does not check consistency of model type and loss selected
"""
function check_args(model::EvoTypes{L}) where {L}
# Check integer parameters
check_parameter(Int, model.max_depth, 1, typemax(Int), :max_depth)
check_parameter(Int, model.nrounds, 0, typemax(Int), :nrounds)
check_parameter(Int, model.nbins, 2, 255, :nbins)
# check positive float parameters
check_parameter(Float64, model.lambda, zero(Float64), typemax(Float64), :lambda)
check_parameter(Float64, model.gamma, zero(Float64), typemax(Float64), :gamma)
check_parameter(Float64, model.min_weight, zero(Float64), typemax(Float64), :min_weight)
# check bounded parameters
check_parameter(Float64, model.alpha, zero(Float64), one(Float64), :alpha)
check_parameter(Float64, model.rowsample, eps(Float64), one(Float64), :rowsample)
check_parameter(Float64, model.colsample, eps(Float64), one(Float64), :colsample)
check_parameter(Float64, model.eta, zero(Float64), typemax(Float64), :eta)
try
tree_type = string(model.tree_type)
@assert tree_type ∈ ["binary", "oblivious"]
catch
error("Invalid input for `tree_type` parameter: `$(model.tree_type)`. Must be of one of `binary` or `oblivious`")
end
end