Home › Community › Crypto-ML Roadmap, Feature, and Suggestions › Poll: Get your vote in. What should be our next ML feature?
- This topic has 16 replies, 10 voices, and was last updated 1 year ago by
admin.
- AuthorPosts
- February 3, 2020 at 3:45 pm #7446
admin
KeymasterPlease take a moment to vote for the machine learning feature you’d most like to see next. What would most help your crypto trading and investing?
Poll here: https://crypto-ml.com/poll-next-machine-learning-project/
Best regards,
Crypto-ML - February 3, 2020 at 7:50 pm #7448
kw
ParticipantI must say I had a really tough time with this poll, as they all sound really useful to me, and they all seem kind of related.
*Bitcoin Dominance Predictor*
Ultimately I voted for this one because I felt it would be the most generally useful to a variety of scenarios.*Signal optimized for short selling*
I’m personally not interested in shorting bitcoin*Signal optimized to maximize BTC value*
I am definitely interested it this. The reason I held back is that I fell this is especially useful during a bear market, and I feel that we are entering (or have already entered) a bull market. After the halving, I will be more interested in maximizing USD instead of BTC. So perhaps we have some runway on this one.*Signal optimized for trading crypto-to-crypto pairs*
Again, I’m definitely interested in this. This would be especially useful in combination withe the dominance predictor. Like right now, BTC dominance is waning, while at the same time BTCUSD has been going up. A great time to trade crypto pairs to gain BTC.*Low-frequency BTC trading signal*
This sounds very useful, but I’m afraid I would not have the patience to see how it would work in live market conditions. - February 3, 2020 at 10:07 pm #7452
Justin
ModeratorGreat write up @kw. Thanks for sharing this detail.
You’ve pretty well summarized the feelings on our end as well. There are a lot of good, related next steps.
Short selling: this is the least exciting one for me personally to get behind. While it makes sense from a trader’s $$ perspective, it feels like rooting for the other team.
Low-frequency: yes…you nailed it. It may take too long to see how it pans out. Also, our research has shown the fewer trades, the more likely you are to miss big movements. This sounds intuitive…but even allowing the math to play out, we just see better results if we don’t constrain frequency. And of course, machines are going to be better at predicting the near term than the long term. That said, I am still optimistic we can build a useful model in this space.
- February 4, 2020 at 4:39 pm #7473
Rob D
ParticipantI didnt vote in the poll because I would love multiple portfolio support for coinbase to do all 4 major signals but maybe thats just me.
- February 4, 2020 at 7:02 pm #7475
Justin
ModeratorHello, thanks for noting that. The portfolio support will be handled by a different team (the API team). The portfolio management feature is still in the pipeline.
The poll was just for the machine learning work, which is handled separately.
- February 4, 2020 at 7:25 pm #7476
kw
ParticipantI was just setting up Coinbase Pro multiple portfolios. They support max 5 portfolios. So if we could set up multiple Auto trades, I could auto trade all 4 pairs!!!
Would be nice workaround (which seems like less effort) until you guys get could get consolidated portfolio management. Food for thought!
- February 5, 2020 at 6:17 am #7477
Bug001
ParticipantBTC value maximization signals will be the most useful for me. Anyway focusing on Dominance first makes more sense.
My personal ranking:
1. Bitcoin Dominance Predictor
2. Signal optimized to maximize BTC value
3. Signal optimized for trading crypto-to-crypto pairs
4. Signal optimized for short selling
5. Low-frequency BTC trading signal - February 5, 2020 at 8:26 am #7478
Justin
Moderator@kw- Yes, this is good and could be an interim workaround. Let us explore this. Thanks for sharing.
@Bug001- thanks for the ranking. The results so far are so close, it’s not giving a super clear picture. I tend to agree with your stack rank there. Items 1, 2, and 3 really go together.
- February 9, 2020 at 3:12 am #7502
cryptologic
ParticipantWhy do people opt for the bitcoin dominance predictor? Can “Bitcoin dominance” be autotraded by CryptoML?
- February 10, 2020 at 11:46 am #7523
Justin
Moderator@cryptologic, the idea is it would help you choose what to focus your trading on–Bitcoin or others.
With Auto Trade as it is today, if you are most bullish on Bitcoin, you have BTC turned on for Auto Trade. But the “Dominance Predictor” could indicate it expects altcoins to rise faster. In that case, you could change your Auto Trade to ETH.
With Auto Trade as it will be in the future, with portfolio management, you may choose to allocate a higher or lower percentage of your trading to Bitcoin vs others based on this mechanism.
- February 10, 2020 at 12:13 pm #7524
martyf
ParticipantI took the poll and I forget what I voted for, but one thing I ponder – will there be continued improvements of the primary BTC ML model?
I know I’m on here frequently venting about lost profits on one hand, and then I sing praises for the bot on the other – but it seems that sometimes the best use of the ML bot is entries…, and you should mix some standard TA for exits.
case in point, this past weekend. CME futures closed on friday at 9850 (they currently have about 100$ premium, spot closed at 9750ish). Bitcoin goes up to 10200 over the weekend, testing the old support of the september meltdown. Weekends are known for looser price movement (not as many pro traders, thus less liquidity). Thus, there’s some n probability that the price is going to come back down to 9750, after there being some x probability that 10200 is gonna have some resistance….
but i know, i know. Bots gonna do what bots gonna do.
Yeah. I guess this rambling is just more to the point that none of the proposed ML projects really buttered my bread, or whatever the saying is. The dominance thing getting the lead is concerning. I don’t have high confidence that a lot of alt coins can maintain truly technically driven price action.
I mean, I know its ridiculous to ask, considering the vast improvements in the 4.0 model, but if I had my druthers the ML team would continue their pursuit of making a model that captures as close to 100% of the gainz as possible.
but i also know that banging your head against the same wall repeatedly isn’t fun.
- February 10, 2020 at 1:12 pm #7525
Justin
Moderator@martyf, I’m glad you asked this. To let the cat out of the bag a bit, on the ML side, we actually already have our 5.x models done. It’s coming fast on the heels of the 4.x models, but it’s a natural evolution.
Getting those deployed, however, is a different animal than in the past. It’ll take next-level systems to operationalize it–which means we need to rework a portion of our infrastructure. Therefore, it’s not on the ML team’s plate; rather, it’s with our API/backend team.
We will have a write-up on it in the next few weeks.
- February 10, 2020 at 6:38 pm #7530
martyf
Participant@Justin – nice.
Always good to know the maths need better hardware!
- February 13, 2020 at 6:23 am #7616
swivelx
ParticipantA bitcoin dominance meter alongside alt-btc trading pairs would be so lit. A lot of alts are popping off against btc and most are outperforming btc by a wide margin. It would be so good if we could capture the riskier, yet highly rewarding trading of altcoins!
- February 19, 2020 at 3:27 am #7658
edbaff1ed
ParticipantIt’s not clear to me the relevance of a bitcoin dominance predictor. If you’re optimising the prediction of bitcoin dominance, that’s what you’ll get – better dominance predictions, which won’t translate necessarily to profits.
- February 21, 2020 at 12:55 am #7676
dannyS
ParticipantCan we get an API notification system, instead of the email notifications?
- February 21, 2020 at 8:53 am #7677
admin
KeymasterHi @dannys, yes API access is coming and is in fact already in alpha testing.
This poll was specifically for the machine learning team. The API/DB team has a different plate of work.
- AuthorPosts
- You must be logged in to reply to this topic.