Machine Learning - Stock Prediction LSTM

sequence_length = 24

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [4299, 1, 24, 5]

ground_truth.shape = [4299, 1, 24, 1]

n_epochs = 34

lr=0.00001

hidden_dim = 120

n_layers = 4

Full dataset normalized

Sigmoid fully connected layer

sequence_length = 24

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [4299, 1, 24, 5]

ground_truth.shape = [4299, 1, 24, 1]

n_epochs = 20

lr=0.00001

hidden_dim = 120

n_layers = 4

Full dataset normalized

no activation function in the fully connected layer

sequence_length = 22

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [4299, 1, 24, 5]

ground_truth.shape = [4299, 1, 24, 1]

n_epochs = 20

lr=0.00001

hidden_dim = 64

n_layers = 4

Full dataset normalized

no activation function in the fully connected layer

sequence_length = 132

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [4299, 1, 24, 5]

ground_truth.shape = [4299, 1, 24, 1]

n_epochs = 100

lr=0.00001

hidden_dim = 8

n_layers = 5

Full dataset normalized

no activation function in the fully connected layer

Same as above trained 200 epochs

Same but 300 epochs

sequence_length = 22

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [4299, 1, 24, 5]

ground_truth.shape = [4299, 1, 24, 1]

n_epochs = 60

lr=0.00001

hidden_dim = 4

n_layers = 3

Full dataset normalized

no activation function in the fully connected layer

sequence_length = 22

batch_size = 1

input = "Open", "High", "Low", "Close", "Volume"

output = Close

train_x.shape = [8260, 1, 24, 5]

ground_truth.shape = [8260, 1, 24, 1]

n_epochs = 60

lr=0.00001

hidden_dim = 32

n_layers = 4

Full dataset normalized

no activation function in the fully connected layer

1 hour data

60 past data 'Close' values

batch size = 1