"Research is creating new knowledge." -Neil Armstrong
Not much has happened, sent an email to the creators of the code that has been the bane of our existence. We await their response. Hopefully everything is sorted out soon its been too stressful honestly.
Also, here is a photo of a good boi splooting
After countless days we finally decided to contact the competitors in order to find a work around. In contacting us back, we were told that we needed to run their code on a machine that has at least 256 GB of RAM and 20 CPU cores.
This was completely crazy since we had no access whatsoever to any kind of machine that has near the level of specs that they recommended.
In the meantime we decided to take a break from running their models and I started on researching on time series data since the LSTM model would require this In order to predict sepsis hours or even minutes before.
The competition code problems still haven’t been resolved. Our supervisor recommended that we should email the creators to see if they could provide assistance. Hopefully they respond soon.
We attempted splitting the files but that didn’t help at all.
Many hours this week was spent at dissecting the competition’s code to verify their results on our datasets, even having sought out the assistance of some Master students and the University who are well versed in neural networks to aid in fixing as well as documenting their code in an attempt to discover what some of their notebooks were for.
Our definition of sepsis, the current sepsis-3 including the being worked upon sepsis-4 definition factors were considered in our choice of features.
We can’t seem to get the competition code to run, it keeps throwing random errors not sure exactly what the problem is. It is so poorly documented and messy. Hopefully we figure it out soon.
Also getting doctors to review the method they utilized to predict sepsis to get feedback so we know what to do in our method of predicting sepsis. Hopefully there is room for improvement so we can achieve better results.
Not much else to report at this time look forward to the next post soon!
Still no word from the physician’s yet so we decided to take another crack at running the competition’s model. Their code mainly failed at the data pre-processing stage.
We received tons of errors and warnings for improper use of pandas methods on the data frames and spent most of our nights Googling how to fix these errors and referring to stack overflow posts to find work-arounds for what they tried to implement in some of their functions.
Trying to understand the inner workings of the models are a bit technical but we are trying our best at it.
So I successfully got sick. I was unable to do anything all weekend. Had a terribly sore throat and fever that would not go away. However, I was all better by Tuesday and was able to continue my readings on ODEs and understanding sepsis to have a better understanding of which patient information would be vital for determining early onset sepsis.
We also spent some time looking at the other teams code to see exactly what they had done. However, we found some very interesting code where they used an If True statement which left us utterly baffled. Apart from this the code in files were overall confusing and hard to navigate. In our initial running of their code it failed since for some reason it said we were missing a column. We have yet to figure out a solution for this problem.
Hopefully things start to go better.
Much time was spent attempting to run the competition’s code. However, after leaving their code running for over an entire day a compilation error over a column name. Stunted progress of running their code for comparison. As looking through their notebooks and code, showed barely any documentation and some utterly confusing logic in their code.
After getting advice from our supervisor we took another shot at attempting to run our competitors code . In this run we actually made it past the point that we were last week but it still wasn’t enough.
Additionally we contacted a physician for assistance in predicting sepsis . This physician assessed the competitors code to see if they were consistent and accurate in their method of predicting sepsis.
We are hoping to create a new method of predicting sepsis using machine learning.
In the mean time we would just continue to research the models.
In addition to researching more neural network models, I’m learning about the recurrent neural networks the most.
This week was mainly about running our competitors code, but in looking at the code repository on GitHub it was very messy and unorganized . A lot of time went into actually ripping apart the Python files and putting them into Jupyter notebooks .
Jupyter notebooks actually helped in providing a means by restarting at a workable point and continuing instead of having to run an entire Python file all over again . There were so many issues that we needed to contact our supervisor to get more advice on how to proceed with this task .
It was necessary since we needed to benchmark all models performances to theirs.
What if they just haven’t made their actual code available ?!
The Stallions Chronicle