[pinouchon]

Deep Learning (by Aaron Courville, Ian Goodfellow, and Yoshua Bengio) Read cover to cover, then applied these techniques to Crypto Markets (turns out, crypto markets are pretty inneficient)

[_9omd]

Interesting. Do your trading strategies still work, or has the market become too efficient? Did you ever trade in a bear market, and by how much did you beat returns on holding just BTC and ETH?

[pinouchon]

We have experience trading since march 2020, so no proper experience in a bear market. We deployed our first algos with nontrivial amounts (> 100usd) the night before the march crash. And yes we did get rekt that night :) Trading strategies work. That it still works is a huge mystery to me. We did beat BTC, we have no long/short bias, and are doing the best during volatile periods (up or down)

[alexmingoia]

You may be interested in earning money from your models with stocks at https://numer.ai, if you don’t already know about it.

[carabiner]

I hate reading comments like this because it always causes me to waste a week dabbling in jupyter notebooks with bitcoin data before realizing it's probably far too much work for me. Happens every 10 months, on average. I know it's 99% data cleansing to arrive at a "true" price through the noise. How to do that, I don't know.

[abrichr]

I assume if you had consistent success you were doing more than time series forecasting on prices. Any chance you can elaborate a bit about what were the model inputs and outputs?

[xcubic]

Did you know ML before that?

[pinouchon]

Before that, I had a strong interest in AI & ML, read the Deep Learning book and did the cs231n mooc. So I could train basic neural networks with pytorch but that was it.