src/callbacks/logging/mlflow.jl: Add basic MLFlowBackend.

Project.toml

@@ -13,6 +13,7 @@ Glob = "c27321d9-0574-5035-807b-f59d2c89b15c"
 Graphs = "86223c79-3864-5bf0-83f7-82e725a168b6"
 ImageCore = "a09fc81d-aa75-5fe9-8630-4744c3626534"
 InlineTest = "bd334432-b1e7-49c7-a2dc-dd9149e4ebd6"
+MLFlowClient = "64a0f543-368b-4a9a-827a-e71edb2a0b83"
 OnlineStats = "a15396b6-48d5-5d58-9928-6d29437db91e"
 Optimisers = "3bd65402-5787-11e9-1adc-39752487f4e2"
 ParameterSchedulers = "d7d3b36b-41b8-4d0d-a2bf-768c6151755e"

src/FluxTraining.jl

@@ -25,6 +25,7 @@ using ChainRulesCore
 import ParameterSchedulers
 import ParameterSchedulers: Sequence, Shifted, Sin
 using TensorBoardLogger: TBLogger, log_value, log_image, log_text, log_histogram, tb_overwrite
+using MLFlowClient
 using Zygote: Grads, gradient
 using ValueHistories
 using DataStructures: DefaultDict, PriorityQueue, enqueue!, dequeue!
@@ -48,6 +49,7 @@ include("./callbacks/execution.jl")
 include("./callbacks/logging/Loggables.jl")
 include("./callbacks/logging/logger.jl")
 include("./callbacks/logging/tensorboard.jl")
+include("./callbacks/logging/mlflow.jl")
 include("./callbacks/logging/checkpointer.jl")
 
 
@@ -111,6 +113,7 @@ export AbstractCallback,
     LogHyperParams,
     LogVisualization,
     TensorBoardBackend,
+    MLFlowBackend,
     StopOnNaNLoss,
     LearningRate,
     throttle,

src/callbacks/logging/mlflow.jl

@@ -0,0 +1,27 @@
+using MLFlowClient
+
+"""
+    MLFlowBackend(tracking_uri, experiment_name[; kwargs...])
+
+MLFlow backend for logging metrics. Creates a new experiment if it doesn't exist
+and starts a new run.
+"""
+struct MLFlowBackend <: LoggerBackend
+    mlf::MLFlowClient.MLFlow
+    experiment::MLFlowClient.MLFlowExperiment
+    run::MLFlowClient.MLFlowRun
+    function MLFlowBackend(tracking_uri::String, experiment_name::String; kwargs...)
+        mlf = MLFlowClient.MLFlow(tracking_uri)
+        experiment = MLFlowClient.getorcreateexperiment(mlf, experiment_name)
+        run = MLFlowClient.createrun(mlf, experiment)
+        return new(mlf, experiment, run)
+    end
+end
+
+Base.show(io::IO, backend::MLFlowBackend) = print(
+    io, "MLFlowBackend(", backend.mlf.apiroot, ", ", backend.experiment.name, ")")
+
+function log_to(backend::MLFlowBackend, value::Loggables.Value, name, step; group = ())
+    name = _combinename(name, group)
+    MLFlowClient.logmetric(backend.mlf, backend.run, name, value.data, step=step)
+end