Predicting the Price of Bitcoin, Intro to LSTM

Here are a few things that jumped out as I read through:

• It's very suspect that your validation loss is consistently lower than your training loss. I would double-check your univariate_data function to make sure training data isn't bleeding into the test set.
• A classic way to graph the loss comparison extends further to the right, to see when validation loss starts moving up (visualize where overfitting occurs). Also, I'd recommend being more explicit about the loss function you use, right now your choice is buried in a block of code - why is MSE right for this use case?
• The number of parameters in your LSTM is much larger than the size of your training set. There are ways to accommodate this (e.g. random dropout) but this would be worth addressing in your article - why is a model this large appropriate for a dataset of daily closing prices spanning 6 years?
• Your historical price graph is flipped along the x-axis, it shows the price plummeting to zero after the 19k peak and staying there to present day
• It looks like large movements in your predictions slightly lag movement in actuals, this would be interesting to discuss in the article in terms of implications for a trading strategy