Hope Alphien team fix the problem

Alphathon
#1

Dear Sir/Madam,


I write this letter to you about the problems I face in this platform.


I have joined different open quant competitions for a long time, such as Quantopian, Quantiacs, FundSeeder…, etc. But this is my first time to know Alphien. And it gives me a bad impression.


I published “If UBS doesn’t allow me to trade ETF, how can I re-simulate ETF portfolio?” dated on EDT 2020/09/14, but this article just disappeared today. The key message of that article is about survivorship bias because only the current trading stocks can be found. This is a HUGE problem. Apart from survivorship bias, there are other problems.


Here is the list:

  1. Memory Problem
  2. Some python package cannot be imported or I don’t know how (I seldom use Jupyter Book). One of the packages which I fail to import is mlfinlab (Installation — mlfinlab 1.0.0 documentation)
  3. Multi-threading is not allowed because of memory problem.
  4. Alternative data is not allowed
  5. Data missing (Delisted stocks)


I would welcome the opportunity to discuss matters further. Hope Alphien fix the above-mentioned problems.


I look forward to hearing from you.


Yours faithfully,



#2

Hi there,

Everyone on the platform has an equal chance of giving his/her solution to a challenge or competition we launch. To maintain that level of fairness across participants, all participants have the same hardware to use (memory in particular, but also CPU), access to the same dataset and functions, and finally access to the same information/documentation.


In that respect, memory is constrained by design. 4Go is plenty of memory to use. Similarly, given we only manipulate tabular data, access to 2 CPUs is also plenty. Given the amount of data at stake, there is no need for multi-threading; it would have too much overhead compared to the benefits.


You can install packages with conda or pip in your environment; this is allowed and explained here: Alphien Dashboard


There is no alternative data allowed as per sponsor’s requirement for this competition. Why don’t you try to build a portfolio with a subsample of available data first? A systematic investment is not great because it uses that much data, multi-threading or a fancy machine learning model, it is only has value when it can generalise, i.e. have good performance on new market data.


Please elaborate on the delisted stock you think are missing. Have you checked the function getTickersSP500() which is in the data package?

alphien.data.getTickersSP500().ticker.shape

(679,)

There are 679 tickers while the S&P has 505 components at present. Where do you think the difference comes from? If you can elaborate your point, we will of course take this into account and add missing ones if any.


Overall, we welcome feedback and we thank you for it. However, please make sure you elaborate of give as much details as possible so that we can improve your experience.



#3

I would say that I can tell the support team itself is helpful and is trying to solve what they can solve, but the overall platform do have plenty of space to improve…


The notebook is really very inconvenient and same code fails from time to time like some random variable even if there are plenty of memory left.

#4

Hello - To answer your question around hope - the answer is YES, we will fix any platform problems that you encounter as long as they are problems that you can’t fix yourself. We hope you can submit a good algorithm for the competition.


It appears you don’t fully understand how to deal with survivorship bias in a stock picking competition, as answered on : https://dashboard.alphien.com/forum/207 regarding delisted stocks. I suggest you ask clarification on that thread if that’s not clear to you. If you have done strategies before on single stock you should know that to avoid survivorship bias you need to include delisted stocks.


Regarding your other points :

1- Memory : Is the issue lack of memory or memory is not working well ? I suggest you post a specific post with your problem (on thread : https://dashboard.alphien.com/forum/213)

2- Python package : let us know which Python package you wish and we will help you import it if you can’t import it in your environment yourself.

3- Multi-threading is not allowed on our platform because of limitation on CPU usage.

4- Alternate Data is allowed : check competition terms to know the data which is allowed to use. Some Macro data and other data is allowed as well as industry classification for stocks. Again this is not an ETF competition (that is easier) as we had in the past.

5- Please let us know which delisted stock for the period of 2007 to 2016 is missing ? Just name one.


@kelvinli - I suggest you look at https://dashboard.alphien.com/forum/208 point 2 : I quote :

"2- The notebook environment is particularly heavy on memory usage, we recommend using notebook only when you have finalized your code to share your code with your team or publicly and submit a notebook. Alphien Lab and Alphien Studio are much ‘faster’ environments to work with"


As Herve mentions, we will support the entire community for competitions, learning how the platform works will help you to win and also learn data science as well.


That being said, again, there are certain constraints of the platform that we are working on to improve the user experience, we are not clear which one is the one we should work on based on your comments.


Happy coding ! Lionel.