fix example

example1.py

@@ -6,22 +6,19 @@
 
 import mfpbench
 
-# Optional imports here just to give typing and so you can look deeper if needed
-from mfpbench.jahs import JAHSBenchmark, JAHSConfig, JAHSResult
-
 if __name__ == "__main__":
-
     seed = 724
 
-    # Just adding a type here if you want to explore it
-    benchmark = mfpbench.get("jahs_cifar10", seed=seed)  # datadir = ...
-    benchmark = cast(JAHSBenchmark, benchmark)
+    # Just adding a type here if you want to explore it # get(..., datadir = ...)
+    benchmark: mfpbench.Benchmark = mfpbench.get("lcbench", task_id="3945", seed=seed)
+
+    min_fidelity, max_fidelity, fidelity_step = benchmark.fidelity_range
 
     # benchmark = mfpbench.get("lcbench", seed=seed, datadir=datadir, task_id="3945")
     # benchmark = cast(LCBench, benchmark)
 
     # Get a random config just to see it
-    config: JAHSConfig = benchmark.sample()
+    config: mfpbench.Config = benchmark.sample()
 
     # And the search space
 
@@ -30,20 +27,21 @@
     assert exact_copy == config
 
     # You can only mutate it by a copy to keep things consistent
-    new_copy = config.copy(TrivialAugment=True)
+    new_copy = config.copy(momentum=0.2)
 
     # You can always validate a config to make sure you aren't doing anything wrong
     new_copy.validate()
 
     # Like in this case where we used a bad optmizer
-    bad_copy = config.copy(Optimizer="Adam")
+    bad_copy = config.copy(momentum=-10)
     try:
         bad_copy.validate()
     except AssertionError:
         pass
 
     # Anyways, here's the results for the config
-    result: JAHSResult = benchmark.query(config, at=42)
+    result: mfpbench.Result
+    result = benchmark.query(config, at=42)
     result = benchmark.query(config.dict(), at=42)  # You can also use a dict
     result = benchmark.query(benchmark.space.sample_configuration())  # Or configspace
 
@@ -53,9 +51,13 @@
     # The full result object
 
     # And if you need the full trajectory
-    results: list[JAHSResult] = benchmark.trajectory(config)
-    sliced_result = benchmark.trajectory(config, to=100)
-    sliced_result_2 = benchmark.trajectory(config, frm=50, to=100)
+    results: list[mfpbench.Result] = benchmark.trajectory(config)
+    sliced_result = benchmark.trajectory(config, to=max_fidelity)
+    sliced_result_2 = benchmark.trajectory(
+        config,
+        frm=max_fidelity // 2,
+        to=max_fidelity,
+    )
 
     first = results[0]
     last = results[-1]
@@ -65,7 +67,7 @@
     best = sorted_result[0]
 
     # Now here's 100 configs and we'll get the best configuration
-    configs: list[JAHSConfig] = benchmark.sample(100)
+    configs: list[mfpbench.Config] = benchmark.sample(100)
 
     # Get all trajectories for each run
     # [

mfpbench/__init__.py

@@ -20,6 +20,7 @@
     PD1translatewmt_xformer_64,
     PD1uniref50_transformer_128,
 )
+from mfpbench.result import Result  # noqa: F401
 from mfpbench.synthetic.hartmann import (
     MFHartmann3Benchmark,
     MFHartmann3BenchmarkBad,

pyproject.toml

@@ -15,6 +15,7 @@ tqdm = "*"
 numpy = "1.*"
 yahpo-gym = "1.0.1"
 xgboost = "^1"
+ConfigSpace = "<0.7"
 
 [tool.poetry.group.jahs.dependencies]
 jahs-bench = { git = "https://github.com/automl/jahs_bench_201.git", rev = "880fbcb35a83df7b6c02440a6c13adb921f54657" }