Comments (7)
@sumanabasu Thank you so much for your comments!
- Re: moving this line to
__init__
, actually this line in__init__
will call_seed()
function. - Re: my plan for generating the new episode, my plan is of course as you said changing those 2 seeds deterministically. Then this start_time will be regenerated. My next plan is to make the initial BG settable (see this issue). And set the initial BG randomly based on the seed.
- Re the patient parameters: the patient parameters are saved in this file. If you want to change the parameters, this is the place to change it.
- Thanks a lot if you want to contribute to a better documentation! Feel free to shoot me PRs.
from simglucose.
Thanks for bringing up the issue. If a seed is set, I don't think reset should reset the seed as well. For example, in this basic environment, reset won't reset the seed. So the example you show here is expected.
But as you said, the environment generates the same episode after changing the seed. This is definitely a bug. I have a temporary fix in my local branch. Will submit the fix soon. Thanks!
from simglucose.
The fix is submitted. Now the results will be like this:
import gym
from gym.envs.registration import register
register(
id='simglucose-adult001-v0',
entry_point='simglucose.envs:T1DSimEnv',
kwargs={'patient_name': 'adult#001'}
)
env = gym.make('simglucose-adult001-v0')
env.seed(0)
print(env.reset()) # >>> Observation(CGM=157.25395877711873)
print(env.reset()) # >>> Observation(CGM=157.25395877711873)
env.seed(1)
print(env.reset()) # >>> Observation(CGM=134.48866253152116)
print(env.reset()) # >>> Observation(CGM=134.48866253152116)
Note that reset()
won't change the random state, if you want to change the random state, use seed(another_seed)
instead.
from simglucose.
Hi, I am not sure about the fix. Can we discuss it?
In env.py.reset()
everything is reset, that is, patient, sensor, pump and scenario.
For patient, it is reset to her initial state. I am not familiar with the UVA/Padova model but I guess this may be a need of the model.
In pump there is no random state.
But for sensor and scenario, in their reset()
methods, the random state set again with the initial seed. So it is generating exactly the same episode and these are the only sources of randomness.
You can see below a snippet from a log:
create_scenario {'meal': {'time': [363.0, 581.0, 803.0, 1158.0, 1340.0], 'amount': [35, 3, 71, 94, 7]}}
_reset BG 149.01999999999998
_reset CGM 159.63251057672602
reset CGM 161.28268202215403
Called reset
Resetting CGM sensor ...
create_scenario {'meal': {'time': [363.0, 581.0, 803.0, 1158.0, 1340.0], 'amount': [35, 3, 71, 94, 7]}}
_reset BG 149.01999999999998
_reset CGM 159.63251057672602
reset CGM 161.28268202215403
Is this the intended behavior? If it actually is, could you explain why?
As I read from the gym documentation, the episode should be independent:
Thanks a lot @jxx123
def reset(self):
"""Resets the environment to an initial state and returns an initial
observation.
Note that this function should not reset the environment's random
number generator(s); random variables in the environment's state should
be sampled independently between multiple calls to `reset()`. In other
words, each call of `reset()` should yield an environment suitable for
a new episode, independent of previous episodes.
Returns:
observation (object): the initial observation.
"""
from simglucose.
@eseglo Thanks for the latest gym document (somehow my old version gym does not document this). I see the problem now. I misunderstood it before. So the requirement is that reset() won't change the seed or random state, but the episode should be different from the previous one after calling reset().
Cool, let me reopen this issue. Will work on a fix soon. Thanks!
from simglucose.
@jxx123 Thank you for trying to fix this! I was using the workaround presented below. Basically, I was changing the seed to the sensor and the scenario every time after I call the reset()
. I thought this generates new episodes as:
- Sensor: It changes the noise generator of the sensor (hence different CGM even if we start from the same BG for a patient)
- Scenario: It changes the seed responsible for creating a new scenario
env = gym.make('simglucose-adult001-v0')
env.reset()
env.env.sensor.seed = seed
env.env.scenario.seed = seed
So, if we set these two seeds inside the old _seed()
function that you had, it looks about right to me. Although, we might want to move this line to the _init_
if we do not want to change the self.np_random
between resets. If I understand @eseglo 's comment, it also conforms to the gym guidelines, unless I'm missing some other random number generator.
I was wondering how else do you plan to generate new episodes without changing these 2 seeds. In my understanding, they are the only sources of randomness in the model outside the random number generators. Could you please explain? It will help expedite my research without having to wait for the fix.
On a different note, for research purposes, I am curious about how you've generated the patient information for this simulator. Is it the same as the UVA Padova in silico patient data? I thought that's private information given UVA Padova is a paid simulator. I'm trying to understand the benefits of having the full UVA Padova matlab simulator and which data sources of this simulator I need to change if I get hold of the UVA Padova.
Thanks again for your time, effort, and enthusiasm!
I might also want to help you build the documentation for the repo since I've been spending quite some time playing with it.
from simglucose.
The fix is in. Will make the initial BG settable in another PR.
from simglucose.
Related Issues (20)
- The reward function encourages short episodes HOT 2
- Units in the Meal Absoption Model HOT 6
- Report function does not work HOT 2
- Controller is not reset during simulation of patients HOT 5
- Scenario with no meal in gym env HOT 2
- How to correctly implement the use of OpenAI Gym - reset() and step() methods are not implemented (NotImplementedError on google Colab)
- Dbar Modification HOT 3
- whether there is paper related to this project or not?
- Input meal size given in (g/min) instead of expected (g)
- Import Error
- Risk index calculation error in report generation
- Risk Index probably incorrect HOT 3
- Does this simulator has the property of time-delay? HOT 1
- Insulin unit error and scenario setting error HOT 3
- pip 0.2.8 broken HOT 2
- Patient interface missing observation, state and t HOT 2
- Clarity on calculation of parameters apha and beta HOT 2
- Patient Parameter Documentation
- Sensor sample_time affects reported CHO and Insulin
- Deprecated distutils in Python v3.12 causes "no module..." exception
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from simglucose.