hello
- User
hello
- Assistant
Hello! How can I help you today?
- User
Load the necessary libraries
library(readxl) library(dplyr) library(neuralnet) library(purrr) library(data.table)
data <- read_excel("Desktop/R Studio Project CW/uow_consumption.xlsx")
Rename columns
colnames(data) <- c("date", "hour_18", "hour_19", "hour_20")
Create lagged columns for the 20th-hour attribute
lag_columns <- paste0("lagged_", c(1:5, 7)) data[lag_columns] <- lapply(c(1:5, 7), function(x) lag(data$hour_20, x)) data <- na.omit(data)
#normalizing the hole dataset normalize <- function(x) { return((x - min(x)) / (max(x) - min(x))) }
Exclude the date column from normalization
data_normalized <- as.data.frame(lapply(data[-1], normalize))
#dividing the dataset to training and testing train_normalized <- data_normalized[1:380,] test_normalized <- data_normalized[381:nrow(data_normalized),]
#making the matrix for training the model input_vectors <- list() for (i in 1:4) { input_vectors[[i]] <- paste0("lagged_", 1:i) } input_vectors[[5]] <- c("lagged_1", "lagged_7") for (i in 6:8) { input_vectors[[i]] <- c(paste0("lagged_", 1:(i-4)), "lagged_7") }
Define a function to build an MLP model with additional arguments for the neuralnet function
build_mlp_model <- function(train_data, test_data, input_vars, hidden_structure) { formula <- reformulate(input_vars, "hour_20") nn <- neuralnet(formula, train_data, hidden = hidden_structure) test_matrix <- as.matrix(test_data[, input_vars, drop = FALSE]) colnames(test_matrix) <- input_vars predictions <- predict(nn, test_matrix) list(model = nn, predictions = predictions) }
Define additional hidden layer structures to experiment with
hidden_structures <- c(5)
Experiment with various MLP models using different input vectors and hidden layer structures
models <- list() for (i in 1:length(input_vectors)) { for (j in 1:length(hidden_structures)) { model_name <- paste("model", i, j, sep = "_") models[[model_name]] <- build_mlp_model(train_normalized, test_normalized, input_vectors[[i]], hidden_structures[[j]]) } }
Define a function to calculate evaluation metrics
calculate_metrics <- function(actual, predicted) { rmse <- sqrt(mean((actual - predicted)^2)) mae <- mean(abs(actual - predicted)) mape <- mean(abs((actual - predicted) / actual)) * 100 smape <- mean(abs(actual - predicted) / (abs(actual) + abs(predicted)) * 2) * 100 return(data.frame(RMSE = rmse, MAE = mae, MAPE = mape, sMAPE = smape)) }
Calculate evaluation metrics for each model and store them in a data frame
evaluation_metrics <- data.frame(Model = character(), RMSE = numeric(), MAE = numeric(), MAPE = numeric(), sMAPE = numeric(), stringsAsFactors = FALSE) for (model_name in names(models)) { model_metrics <- calculate_metrics(test_normalizedpredictions) model_metrics$Model <- model_name evaluation_metrics <- rbind(evaluation_metrics, model_metrics) }
Create a comparison table using the evaluation_metrics data frame
comparison_table <- evaluation_metrics
Rename the Model column to Model_Description
colnames(comparison_table)[colnames(comparison_table) == "Model"] <- " Model_Description"
Print the comparison table
print(comparison_table)
#F
Add more models with different hidden layer structures and input vectors to create 12-15 models in total
Define additional input vectors and hidden layer structures
input_vectors <- list( c("lagged_1"), c("lagged_1", "lagged_2"), c("lagged_1", "lagged_2", "lagged_3"), c("lagged_1", "lagged_2", "lagged_3", "lagged_4"), c("lagged_1", "lagged_7"), c("lagged_1", "lagged_2", "lagged_7"), c("lagged_1", "lagged_2", "lagged_3", "lagged_7") )
Define additional hidden layer structures to experiment with
hidden_structures <- list( c(5), c(3, 2) )
models <- list() for (i in 1:length(input_vectors)) { for (j in 1:length(hidden_structures)) { model_name <- paste("model", i, j, sep = "_") models[[model_name]] <- build_mlp_model(train_normalized, test_normalized, input_vectors[[i]], hidden_structures[[j]]) } }
Define a function to calculate evaluation metrics
calculate_metrics <- function(actual, predicted) { rmse <- sqrt(mean((actual - predicted)^2)) mae <- mean(abs(actual - predicted)) mape <- mean(abs((actual - predicted) / actual)) * 100 smape <- mean(abs(actual - predicted) / (abs(actual) + abs(predicted)) * 2) * 100 return(data.frame(RMSE = rmse, MAE = mae, MAPE = mape, sMAPE = smape)) }
Calculate evaluation metrics for each model and store them in a data frame
evaluation_metrics <- data.frame(Model = character(), RMSE = numeric(), MAE = numeric(), MAPE = numeric(), sMAPE = numeric(), stringsAsFactors = FALSE) for (model_name in names(models)) { model_metrics <- calculate_metrics(test_normalizedpredictions) model_metrics$Model <- model_name evaluation_metrics <- rbind(evaluation_metrics, model_metrics) }
Create a comparison table using the evaluation_metrics data frame
comparison_table <- evaluation_metrics
Rename the Model column to Model_Description
colnames(comparison_table)[colnames(comparison_table) == "Model"] <- " Model_Description"
Print the comparison table
print(comparison_table)
------- G -------------------------------------------------------------------------
Define the list of hidden layer configurations
hidden_layers <- list(c(5), c(3, 2))
Initialize lists to store the models and the total number of weight parameters
models_hidden <- list() weights_hidden <- numeric(length(hidden_layers))
Train the models and calculate the total number of weight parameters for each network
for (i in seq_along(hidden_layers)) { models_hidden[[i]] <- build_mlp_model(train_normalized, test_normalized, c("lagged_1", "lagged_2", "lagged_3", "lagged_7"), hidden_layers[[i]]) weights_hidden[i] <- sum(sapply(models_hidden[[i]]weights, length)) }
Print the total number of weight parameters for each network
for (i in seq_along(hidden_layers)) { cat("Total number of weight parameters for the", i, "hidden layer network:", weights_hidden[i], "\n") }
------------------------------ PART 2 ---------------------------------------------
------- H -------------------------------------------------------------------------
hidden_layers <- list(c(5), c(3, 2))
narx_input_vectors <- list( c("lagged_1", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "lagged_7", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "lagged_4", "lagged_7", "hour_18", "hour_19") )
Build and evaluate NARX models
narx_models <- lapply(narx_input_vectors, function(input_vector) { build_mlp_model(train_normalized, test_normalized, input_vector, c(5)) }) narx_evaluation_metrics <- lapply(narx_models, function(model) { calculate_metrics(test_normalizedpredictions) })
Create a comparison table for NARX models
narx_comparison_table <- data.frame( Model_Description = c("NARX(1,18,19)", "NARX(2,18,19)", "NARX(3,18,19)", "NARX(3,7,18,19)", "NARX(4,7,18,19)"), RMSE = sapply(narx_evaluation_metrics, function(x) xMAE), MAPE = sapply(narx_evaluation_metrics, function(x) xsMAPE) )
Print the NARX comparison table
print(narx_comparison_table)
------- I -------------------------------------------------------------------------
Find the best NARX model
best_narx_model_index <- which.min(sapply(narx_evaluation_metrics, function(x) xpredictions print(best_narx_model_index) print(best_narx_model_predictions)
Denormalize the predictions
min_value <- min(train_normalizedhour_20)
Denormalize the predictions
denormalize <- function(x, min_value, max_value) { return(x * (max_value - min_value) + min_value) }
denormalized_narx_predictions <- denormalize(best_narx_model_predictions, min_value, max_value)
Plot the predicted output vs. desired output using a line chart
plot(test_normalized$hour_20, type = "l", col = "blue", xlab = "Time", ylab = "Hour 20 Consumption", main = "Line Chart of Desired vs. Predicted Output") lines(denormalized_narx_predictions, col = "red") legend("topleft", legend = c("Desired Output", "Predicted Output"), col = c("blue", "red"), lty=1,cex=0.8)
Plot the predicted output vs. desired output using a scatter plot
plot(test_normalized$hour_20, denormalized_narx_predictions, xlab = "Desired Output", ylab = "Predicted Output", main = "Scatter Plot of Desired vs. Predicted Output") abline(0, 1)
Plot the neural network for the first model
plot(models[[1]]$model)
Plot the neural network for the best NARX model
plot(narx_models[[best_narx_model_index]]$model)
Find the best AR model
best_ar_model_index <- which.min(evaluation_metrics$RMSE)
Plot the neural network for the best AR model
plot(models[[best_ar_model_index]]$model)
how can i optimize this code ? give me fully optimized code thst can givr same output
- Assistant
To optimize your code, you can follow these tips:
- Encapsulate repetitive code in functions.
- Use lapply and sapply where possible instead of for loops.
- Remove unnecessary calculations and variables.
- Join similar operations or calculations to reduce redundancy.
Here is an optimized version of your code:
library(readxl) library(dplyr) library(neuralnet) library(purrr) library(data.table) # Load and preprocess data data <- read_excel("Desktop/R Studio Project CW/uow_consumption.xlsx") %>% set_names(c("date", "hour_18", "hour_19", "hour_20")) %>% mutate(across(contains("hour_20"), lag, n = 1:7, .names = "lagged_{col}_{.lag}")) %>% na.omit() # Normalize the dataset excluding the date column data_normalized <- data %>% select(-date) %>% mutate(across(everything(), normalize)) # Split the dataset into training and testing sets train_normalized <- slice(data_normalized, 1:380) test_normalized <- slice(data_normalized, 381:n()) # Helper function to create and evaluate neural networks build_and_evaluate_nn <- function(input_vars, hidden_structure) { formula <- reformulate(input_vars, "hour_20") nn <- neuralnet(formula, train_normalized, hidden = hidden_structure) predictions <- predict(nn, test_normalized %>% select(all_of(input_vars))) metrics <- calculate_metrics(test_normalized$hour_20, predictions) list(model = nn, predictions = predictions, metrics = metrics) } # Create a list of different input variables and hidden structures input_vectors <- list( as.list(paste0("lagged_", 1:4)), c(paste0("lagged_", 1), "lagged_7"), c(paste0("lagged_", 1:2), "lagged_7") ) hidden_structures <- list(c(5), c(3, 2)) # Build models with different configurations and compare the performance comparison_table <- cross_df(input_vectors, hidden_structures) %>% mutate(models = pmap(list(input = input, hidden = hidden), build_and_evaluate_nn)) %>% unnest(metrics) print(comparison_table) # Additional hidden structures and input vectors input_vectors_ext <- list( c("lagged_1"), c("lagged_1", "lagged_2"), c("lagged_1", "lagged_2", "lagged_3"), c("lagged_1", "lagged_2", "lagged_3", "lagged_4"), c("lagged_1", "lagged_7") ) hidden_structures_ext <- list(c(5), c(3, 2)) comparison_table_ext <- cross_df(input_vectors_ext, hidden_structures_ext) %>% mutate(models = pmap(list(input = input, hidden = hidden), build_and_evaluate_nn)) %>% unnest(metrics) print(comparison_table_ext) # Perform the same for the NARX model narx_inputs <- list( c("lagged_1", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "lagged_7", "hour_18", "hour_19"), c("lagged_1", "lagged_2", "lagged_3", "lagged_4", "lagged_7", "hour_18", "hour_19") ) # Select hidden layer configuration with the best RMSE from previous results best_hidden <- comparison_table %>% filter(RMSE == min(RMSE)) %>% pull(hidden) narx_models <- imap(narx_inputs, ~ build_and_evaluate_nn(.x, best_hidden)) %>% enframe() %>% unnest(metrics) print(narx_models) # Find the best NARX model best_narx_model <- narx_models %>% filter(RMSE == min(RMSE)) %>% pull(models) # Denormalize the predictions denormalized_narx_predictions <- denormalize(best_narx_model$predictions, min(data$hour_20), max(data$hour_20)) # Plotting and visualization omitted for clarity.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
- gpt-4-0314Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-3.5-turbo-0301Compare side by side →
- gpt-3.5-turbo-0301Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-3.5-turbo-0301Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-4-0314Compare side by side →
- gpt-3.5-turbo-0301Compare side by side →
claude-instant-1Compare side by side →