Finding a clever way to set a threshold given a list of loss values
Assume I have a list of losses plotted in the following KDE plot:
If the goal is to spot the outliers, the best threshold would be clearly around 0.75, where the value of density is the minimum possible (near zero), and it is at the beginning of the tail.
Given a list of loss values, how can I (as accurately as possible) set such a threshold at the beginning of the tail?
Solution
I would suggest to sort the list of loss values.
Then it depends upon the number of values present.
Suppose there are 100 values in the list and you need to find the threshold up to 95% accuracy, you can decide the threshold value after sorting the list and see beyond which point the last 5 values fall.
- Numerically stable softmax
- ModuleNotFoundError, even after pip install
- how to extract the texts after the first h1 Tag?
- Requests and BeautifulSoup to get video length from YouTube
- Python Multiple Turtles Moving (seemingly) Simultaneously
- Stop pyzmq receiver by KeyboardInterrupt
- pandas dataframe update with filter_func
- Getting the name of a variable as a string
- using bottlenose in python to extract product price from Amazon in different locale
- matplotlib multiple axes mixups
- ZMQ socket gracefully termination in Python
- gdown command not working in jupyter notebook in anaconda.cloud
- Can dependency injection be used with django?
- Simple way to measure cell execution time in ipython notebook
- Asyncio future, running Future objects
- Prime factorization using list comprehension in Python
- Store output of subprocess.Popen call in a string
- sqlalchemy.exc.NoReferencedTableError: Foreign key associated with column could not find table
- How to get the Exception (Message and StackStrace) of a Python Azure Function
- Python - regex to grab specific lines from text
- python: How to get real feature name from feature_importances
- Python: Implement a PCA using SVD
- Apply function only on slice of array under jit
- symbol not found in flat namespace '_bcp_batch'
- Print function not working without flush=True when used with the While loop
- SciPy minimise to find inverse function?
- Microservices architecture with Django
- how to escape texts for formatting in python
- How should I read a file line-by-line in Python?
- Sort columns of numpy unitary matrix such that highest element of each column is on the diagonal