kartik-hegde / mindmappings Goto Github PK

Dear Kartik:
Thank you for your promising work. We've encountered some issues while attempting to replicate the results from your paper. Could you assist us? While reproducing the experiments, we've noticed that the individual optimization process exhibits a spiky result, and we haven't been able to identify a clear convergence trend, as below:

Could you please review whether the code in this repository is the most up-to-date version of your experiments and if the hyperparameters are appropriately set, such as the 'injection_intervals' parameter?

Additionally, we've found a few minor code issues, e.g.,

`
best_cost = next_cost if(next_cost < best_cost) else best_cost

best_mapping = projected_mapping if(next_cost < best_cost) else best_mapping
`

Your help would be greatly appreciated.

Hi,
We found a possible for provided surrogate model, that when we run reproduce command, we notice that the number of EDP Normalized to Alg. Min is nearly 5, but when run search, the searched best mapping's cost is over 100 when normalized to Alg. Min.

The figure below illustrates our doubt.

We don't know if we use a wrong model or we do some mistakes.

Just correct me if I overlooked something. Thank you in advance!

Best regards,
Linyan

Dear authors, thanks for this tool! However, when I followed the setup instructions and executed python3 costModel/timeloop/model_timeloop.py, it reported the EDP to be inf as below, is it an expected (or possible) outputs ? If not, how could this be fixed ?

Hi,

After installing Timeloop, I executed the python3 costModel/timeloop/model_timeloop.py. However, I have noticed that the code is not working as expected, it threw ValueError: could not convert string to float: 'EDP(J*cycle):. And upon further inspection, I found that the positions of the data being parsed maybe swapped.

During debugging, I found the relative position of data is:

While the relevant lines of the code in mindmappings/costModel/timeloop/model_cnn.py#L250-L252:

cost.append(float(data[-24].split(" ")[-1]))
cost.append(float(data[-22].split(" ")[-2]))
cost.append(float(data[-23].split(" ")[-1]))

It seems that these three index number should be continuous. So I swaped the -22 and -23, and python did not raise error then.

Is it really a bug or do I miss anything?