[tune] when using the tune.qrandint, I can not get the upper bound in tune.run and I found that the value range is out of bounds.

[LinWencong]

What happened + What you expected to happen

when i run :

from ray import tune

# This will automatically use the `BasicVariantGenerator`
tune.run(
    lambda config: print(config["a"],config["b"]),
    config={
        "a": tune.qrandint(1, 10, 3),
	    "b": tune.qrandint(1, 5, q=1),
    },
    num_samples=50)

i got following:
+--------------------+------------+-----------------+-----+-----+
| Trial name | status | loc | a | b |
|--------------------+------------+-----------------+-----+-----|
| lambda_87b36_00000 | TERMINATED | 127.0.0.1:96387 | 6 | 2 |
| lambda_87b36_00001 | TERMINATED | 127.0.0.1:96390 | 3 | 2 |
| lambda_87b36_00002 | TERMINATED | 127.0.0.1:96391 | 0 | 2 |
| lambda_87b36_00003 | TERMINATED | 127.0.0.1:96392 | 3 | 4 |
| lambda_87b36_00004 | TERMINATED | 127.0.0.1:96393 | 3 | 1 |
| lambda_87b36_00005 | TERMINATED | 127.0.0.1:96394 | 6 | 1 |
| lambda_87b36_00006 | TERMINATED | 127.0.0.1:96395 | 6 | 4 |
| lambda_87b36_00007 | TERMINATED | 127.0.0.1:96396 | 6 | 1 |
| lambda_87b36_00008 | TERMINATED | 127.0.0.1:96387 | 3 | 1 |
| lambda_87b36_00009 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
| lambda_87b36_00010 | TERMINATED | 127.0.0.1:96387 | 3 | 2 |
| lambda_87b36_00011 | TERMINATED | 127.0.0.1:96387 | 9 | 1 |
| lambda_87b36_00012 | TERMINATED | 127.0.0.1:96387 | 9 | 4 |
| lambda_87b36_00013 | TERMINATED | 127.0.0.1:96387 | 9 | 2 |
| lambda_87b36_00014 | TERMINATED | 127.0.0.1:96387 | 0 | 3 |
| lambda_87b36_00015 | TERMINATED | 127.0.0.1:96387 | 3 | 2 |
| lambda_87b36_00016 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
| lambda_87b36_00017 | TERMINATED | 127.0.0.1:96387 | 9 | 2 |
| lambda_87b36_00018 | TERMINATED | 127.0.0.1:96387 | 6 | 2 |
| lambda_87b36_00019 | TERMINATED | 127.0.0.1:96387 | 6 | 2 |
| lambda_87b36_00020 | TERMINATED | 127.0.0.1:96387 | 6 | 3 |
| lambda_87b36_00021 | TERMINATED | 127.0.0.1:96387 | 3 | 4 |
| lambda_87b36_00022 | TERMINATED | 127.0.0.1:96387 | 9 | 3 |
| lambda_87b36_00023 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
| lambda_87b36_00024 | TERMINATED | 127.0.0.1:96387 | 6 | 1 |
| lambda_87b36_00025 | TERMINATED | 127.0.0.1:96387 | 3 | 4 |
| lambda_87b36_00026 | TERMINATED | 127.0.0.1:96387 | 0 | 2 |
| lambda_87b36_00027 | TERMINATED | 127.0.0.1:96387 | 6 | 1 |
| lambda_87b36_00028 | TERMINATED | 127.0.0.1:96387 | 3 | 4 |
| lambda_87b36_00029 | TERMINATED | 127.0.0.1:96387 | 3 | 2 |
| lambda_87b36_00030 | TERMINATED | 127.0.0.1:96387 | 3 | 1 |
| lambda_87b36_00031 | TERMINATED | 127.0.0.1:96387 | 9 | 1 |
| lambda_87b36_00032 | TERMINATED | 127.0.0.1:96387 | 9 | 3 |
| lambda_87b36_00033 | TERMINATED | 127.0.0.1:96387 | 3 | 4 |
| lambda_87b36_00034 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
| lambda_87b36_00035 | TERMINATED | 127.0.0.1:96387 | 3 | 1 |
| lambda_87b36_00036 | TERMINATED | 127.0.0.1:96387 | 9 | 1 |
| lambda_87b36_00037 | TERMINATED | 127.0.0.1:96387 | 0 | 2 |
| lambda_87b36_00038 | TERMINATED | 127.0.0.1:96387 | 6 | 2 |
| lambda_87b36_00039 | TERMINATED | 127.0.0.1:96387 | 0 | 3 |
| lambda_87b36_00040 | TERMINATED | 127.0.0.1:96387 | 3 | 3 |
| lambda_87b36_00041 | TERMINATED | 127.0.0.1:96387 | 6 | 1 |
| lambda_87b36_00042 | TERMINATED | 127.0.0.1:96387 | 9 | 2 |
| lambda_87b36_00043 | TERMINATED | 127.0.0.1:96387 | 6 | 3 |
| lambda_87b36_00044 | TERMINATED | 127.0.0.1:96387 | 3 | 4 |
| lambda_87b36_00045 | TERMINATED | 127.0.0.1:96387 | 6 | 3 |
| lambda_87b36_00046 | TERMINATED | 127.0.0.1:96387 | 0 | 2 |
| lambda_87b36_00047 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
| lambda_87b36_00048 | TERMINATED | 127.0.0.1:96387 | 3 | 1 |
| lambda_87b36_00049 | TERMINATED | 127.0.0.1:96387 | 6 | 4 |
+--------------------+------------+-----------------+-----+-----+

I have doubts about ：

1. “a” can be 0 ("a": tune.qrandint(1, 10, 3))
2. “b” can not reach 5 (tune.qrandint(1, 5, q=1))

Versions / Dependencies

python = 3.9.7
ray==1.13.0

Reproduction script

from ray import tune

This will automatically use the BasicVariantGenerator

tune.run(
lambda config: print(config["a"],config["b"]),
config={
"a": tune.qrandint(1, 10, 3),
"b": tune.qrandint(1, 5, q=1),
},
num_samples=50)

Issue Severity

No response

[LinWencong]

sample.py line 521

        ...
        values = self.sampler.sample(domain, spec, size, random_state=random_state)
        quantized = np.round(np.divide(values, self.q)) * self.q
        ...

The specific implementation is as above.
Would you mind telling me if this needs to be explained in the documentation; or if there is a bug in the documentation? 😁

[krfricke]

Yes, this looks like a bug. Specifically for a) np.round(1/3) = 0 and b) np.round(4 / 1) * 1 = 4 so in these cases the limits are not correct.

Let's test for these cases in #28187

[xwjiang2010]

cc @ilee300a
sorry I couldn't find a way to assign this to you somehow ...