-
@jstutters as I understand the 250 samples in the prolific study are randomly selected from the 606 original, but will they be balanced so that all samples are named an equal number of times?
For …
-
**Github username:** --
**Twitter username:** --
**Submission hash (on-chain):** 0xdd58e018f662ddffafd4671d664ac3777a65d43a1a96e3c4a2e14d6bc9c058d0
**Severity:** medium
**Description:**
**Descriptio…
-
Hello, I am very interested in your work! I would like to ask about the seed setting for USPS and FASHION data sets. I want to follow your work. Is it convenient to provide the result file?
In the pa…
-
The Frogman (level 70) drops the same type and amount of loot as the opposite karma Stone Troll (level 120).
Loot dropped is a large amount of expensive reagents found on the Island.
For context: Th…
-
Hi,
I have two questions:
First : Why didn't you use K Fold Cross Validation?
Second : What is the reason use different learning rate for classifier? Is it for faster convergence?
I am tryin…
-
as you note,
dataset is imbalance ,so how to use 'happy' data tiwice
-
**Use case**
A client can send many queries over a small number of connections. When the service is behind a proxy and represented by several replicas, the load will be unevenly distributed among t…
-
Using a Cost Matrix will improve the efficiency, especially in case of class imbalance.
-
cc @timhoffm per recent discussion.
-
leader speed is not considered at all