Agree, with one exception: Machine learning with neural networks. In this case you define the features that make up price action, but you let the Neural Network figure out what price action actually means. But, Python is good at that too.
Not quite. My current use case for Neural Networks is where I already know the price action I'm looking for, and it is easier to identify using machine learning pattern recognition than with traditional if/then statements. But I'm working on it.
The basics: I use traditional logic to identify areas on the chart where I'm interested in price action. For example, previous highs/lows, break up/break down candles, gaps. I then use traditional logic to identify when price is within the identified area. At this point, I'll feed real-time price action into the Neural Network, and it will respond with a classification, i.e. predicted bullish or bearish outcome.
If you have predictive features and the neural network is doing a supervised learning classification task then it just a classification task. To say the neural network is figuring out price action is disingenuous and a highly distorted view to someone who doesn't have any experience in ML.
I don't like the absolute specificity that programmers and quants insist on living by. I understood what he meant, it's enough. Flexibility is necessary to have conversations.