Prompt - Task

Construye un modelo LSTM vanilla en TensorFlow/Keras para predecir el número de usuarios diarios de un sitio web, carga el dataset desde 'https://github.com/marsgr6/r-scripts/blob/master/data/viz_data/visitas_por_dia_web_cienciadedatos.csv' usando pandas con parse_dates en la columna 'date' y dayfirst=True, usa la columna 'users' como variable objetivo, realiza análisis exploratorio mostrando la serie de tiempo completa, crea secuencias temporales con lookback window de 30 días, divide los datos respetando el orden temporal (80% de las primeras observaciones para entrenamiento, 20% de las últimas observaciones para prueba), normaliza los datos con MinMaxScaler, construye una arquitectura LSTM con capas [LSTM(50 unidades, return_sequences=False), Dropout(0.2), Dense(1)], compila con optimizer Adam y loss MSE, entrena el modelo con early stopping y validation_split=0.1, realiza predicciones en el conjunto de prueba, desnormaliza las predicciones y valores reales, calcula métricas de evaluación (MSE, RMSE, MAE, MAPE), y genera tres gráficos: (1) serie de tiempo completa mostrando train/test split con línea vertical divisoria, (2) zoom del período de prueba con valores observados vs predichos incluyendo fechas en el eje x, y (3) curvas de pérdida durante el entrenamiento (training y validation loss).

• Nota: Es posible que tenga que insistir a Gemini en la ejecución de las taras. Dividir el prompt en diferentes elementos puede funcionar mejor.

Cargar el dataset

Subtask:

Cargar el dataset desde la URL proporcionada en GitHub usando pandas, asegurando el parseo correcto de la columna 'date'.

Reasoning: Import pandas, define the URL, load the data using pd.read_csv with date parsing, and display the head of the dataframe to verify.

import pandas as pd

url = 'https://github.com/marsgr6/r-scripts/raw/master/data/viz_data/visitas_por_dia_web_cienciadedatos.csv'
df = pd.read_csv(url, parse_dates=['date'], dayfirst=True)
display(df.head())

/tmp/ipython-input-2322786964.py:4: UserWarning: Could not infer format, so each element will be parsed individually, falling back to `dateutil`. To ensure parsing is consistent and as-expected, please specify a format.
  df = pd.read_csv(url, parse_dates=['date'], dayfirst=True)

	date	users
0	2020-07-01	2324
1	2020-07-02	2201
2	2020-07-03	2146
3	2020-07-04	1666
4	2020-07-05	1433

Eda

Subtask:

Mostrar la serie de tiempo completa para visualizar los datos y entender su comportamiento.

Reasoning: Plot the 'users' column against the 'date' column to visualize the time series data and its behavior.

import matplotlib.pyplot as plt

plt.figure(figsize=(12, 6))
plt.plot(df['date'], df['users'])
plt.xlabel('Date')
plt.ylabel('Number of Users')
plt.title('Daily Website Users Over Time')
plt.show()

Preprocesamiento de datos

Subtask:

Seleccionar la columna 'users' como variable objetivo, normalizar los datos utilizando MinMaxScaler y crear secuencias temporales con un lookback window de 30 días.

Reasoning: Import MinMaxScaler, select the target variable, normalize it, and create sequences using a helper function.

from sklearn.preprocessing import MinMaxScaler
import numpy as np

# Select the 'users' column and reshape
target_data = df['users'].values.reshape(-1, 1)

# Initialize and fit the scaler
scaler = MinMaxScaler(feature_range=(0, 1))
scaled_data = scaler.fit_transform(target_data)

# Define the sequence creation function
def create_sequences(data, lookback):
    X, y = [], []
    for i in range(len(data) - lookback):
        X.append(data[i:(i + lookback), 0])
        y.append(data[i + lookback, 0])
    return np.array(X), np.array(y)

# Create sequences with a lookback window of 30
lookback_window = 7
X, y = create_sequences(scaled_data, lookback_window)

print(f"Shape of input sequences (X): {X.shape}")
print(f"Shape of output values (y): {y.shape}")

Shape of input sequences (X): (414, 7)
Shape of output values (y): (414,)

División de datos

Subtask:

Dividir los datos en conjuntos de entrenamiento (80% inicial) y prueba (20% final), manteniendo el orden temporal.

Reasoning: Determine the split index and divide the data into training and testing sets based on the 80/20 ratio while maintaining temporal order.

# Determine the split index (80% for training, 20% for testing)
split_index = int(len(X) * 0.8)

# Split the data into training and testing sets
X_train, X_test = X[:split_index], X[split_index:]
y_train, y_test = y[:split_index], y[split_index:]

# Print the shapes of the training and testing sets
print(f"Shape of X_train: {X_train.shape}")
print(f"Shape of y_train: {y_train.shape}")
print(f"Shape of X_test: {X_test.shape}")
print(f"Shape of y_test: {y_test.shape}")

Shape of X_train: (331, 7)
Shape of y_train: (331,)
Shape of X_test: (83, 7)
Shape of y_test: (83,)

Construir el modelo LSTM

Subtask:

Definir la arquitectura del modelo LSTM en TensorFlow/Keras con capas [LSTM(50 unidades, return_sequences=False), Dropout(0.2), Dense(1)].

Reasoning: Import necessary libraries from TensorFlow and Keras, define the sequential model, add the LSTM layer with 50 units and return_sequences=False, add a Dropout layer with a rate of 0.2, and finally add a Dense layer with 1 unit for the output.

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dropout, Dense

# Build the LSTM model
model = Sequential()
model.add(LSTM(units=50, return_sequences=False, input_shape=(X_train.shape[1], 1)))
model.add(Dropout(0.2))
model.add(Dense(units=1))

# Print the model summary
model.summary()

/usr/local/lib/python3.12/dist-packages/keras/src/layers/rnn/rnn.py:199: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(**kwargs)

Model: "sequential_1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ lstm_1 (LSTM)                   │ (None, 50)             │        10,400 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_1 (Dropout)             │ (None, 50)             │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 1)              │            51 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 10,451 (40.82 KB)

 Trainable params: 10,451 (40.82 KB)

 Non-trainable params: 0 (0.00 B)

Compilar el modelo

Subtask:

Configurar el modelo con el optimizador Adam y la función de pérdida MSE.

Reasoning: Compile the model using the Adam optimizer and Mean Squared Error (MSE) as the loss function.

# Compile the model
model.compile(optimizer='adam', loss='mse')

# Print a confirmation message
print("Model compiled successfully!")

Model compiled successfully!

Entrenar el modelo

Subtask:

Entrenar el modelo utilizando el conjunto de entrenamiento, implementando Early Stopping y validation_split=0.1.

Reasoning: Import EarlyStopping, define the early stopping callback, and fit the model to the training data with the specified parameters.

from tensorflow.keras.callbacks import EarlyStopping

# Define the Early Stopping callback
early_stopping = EarlyStopping(monitor='val_loss', patience=10, restore_best_weights=True)

# Train the model
history = model.fit(
    X_train, y_train,
    epochs=100,  # You can increase the number of epochs, Early Stopping will stop training when necessary
    batch_size=32,
    validation_split=0.1,
    callbacks=[early_stopping]
)

print("Model training finished.")

Epoch 1/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 4s 54ms/step - loss: 0.0850 - val_loss: 0.0234
Epoch 2/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0158 - val_loss: 0.0138
Epoch 3/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0160 - val_loss: 0.0094
Epoch 4/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - loss: 0.0140 - val_loss: 0.0126
Epoch 5/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0124 - val_loss: 0.0084
Epoch 6/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0129 - val_loss: 0.0087
Epoch 7/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0099 - val_loss: 0.0094
Epoch 8/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0105 - val_loss: 0.0087
Epoch 9/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0117 - val_loss: 0.0085
Epoch 10/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0102 - val_loss: 0.0085
Epoch 11/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0098 - val_loss: 0.0089
Epoch 12/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0096 - val_loss: 0.0085
Epoch 13/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0102 - val_loss: 0.0085
Epoch 14/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0113 - val_loss: 0.0095
Epoch 15/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - loss: 0.0096 - val_loss: 0.0083
Epoch 16/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0106 - val_loss: 0.0088
Epoch 17/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0095 - val_loss: 0.0084
Epoch 18/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0111 - val_loss: 0.0083
Epoch 19/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0107 - val_loss: 0.0085
Epoch 20/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0093 - val_loss: 0.0082
Epoch 21/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - loss: 0.0113 - val_loss: 0.0083
Epoch 22/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0111 - val_loss: 0.0082
Epoch 23/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0089 - val_loss: 0.0082
Epoch 24/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0099 - val_loss: 0.0083
Epoch 25/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0111 - val_loss: 0.0080
Epoch 26/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0099 - val_loss: 0.0085
Epoch 27/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0099 - val_loss: 0.0081
Epoch 28/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0110 - val_loss: 0.0083
Epoch 29/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 14ms/step - loss: 0.0106 - val_loss: 0.0080
Epoch 30/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0116 - val_loss: 0.0085
Epoch 31/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0107 - val_loss: 0.0079
Epoch 32/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0089 - val_loss: 0.0080
Epoch 33/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0100 - val_loss: 0.0080
Epoch 34/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0087 - val_loss: 0.0081
Epoch 35/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0083 - val_loss: 0.0077
Epoch 36/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0109 - val_loss: 0.0077
Epoch 37/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0090 - val_loss: 0.0077
Epoch 38/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0100 - val_loss: 0.0076
Epoch 39/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0101 - val_loss: 0.0075
Epoch 40/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0094 - val_loss: 0.0076
Epoch 41/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0105 - val_loss: 0.0075
Epoch 42/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0100 - val_loss: 0.0083
Epoch 43/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0100 - val_loss: 0.0074
Epoch 44/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - loss: 0.0090 - val_loss: 0.0074
Epoch 45/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0095 - val_loss: 0.0072
Epoch 46/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0082 - val_loss: 0.0071
Epoch 47/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0087 - val_loss: 0.0070
Epoch 48/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0088 - val_loss: 0.0071
Epoch 49/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0091 - val_loss: 0.0070
Epoch 50/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0100 - val_loss: 0.0082
Epoch 51/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0108 - val_loss: 0.0078
Epoch 52/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0097 - val_loss: 0.0071
Epoch 53/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0080 - val_loss: 0.0067
Epoch 54/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0095 - val_loss: 0.0066
Epoch 55/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0087 - val_loss: 0.0081
Epoch 56/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0090 - val_loss: 0.0065
Epoch 57/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0081 - val_loss: 0.0065
Epoch 58/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 14ms/step - loss: 0.0098 - val_loss: 0.0064
Epoch 59/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0078 - val_loss: 0.0062
Epoch 60/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - loss: 0.0089 - val_loss: 0.0061
Epoch 61/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0090 - val_loss: 0.0061
Epoch 62/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0080 - val_loss: 0.0085
Epoch 63/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0108 - val_loss: 0.0064
Epoch 64/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0091 - val_loss: 0.0071
Epoch 65/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - loss: 0.0089 - val_loss: 0.0065
Epoch 66/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0082 - val_loss: 0.0063
Epoch 67/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0093 - val_loss: 0.0063
Epoch 68/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0087 - val_loss: 0.0062
Epoch 69/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0083 - val_loss: 0.0067
Epoch 70/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - loss: 0.0079 - val_loss: 0.0058
Epoch 71/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 23ms/step - loss: 0.0091 - val_loss: 0.0057
Epoch 72/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - loss: 0.0089 - val_loss: 0.0053
Epoch 73/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - loss: 0.0089 - val_loss: 0.0058
Epoch 74/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - loss: 0.0077 - val_loss: 0.0054
Epoch 75/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - loss: 0.0078 - val_loss: 0.0052
Epoch 76/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step - loss: 0.0068 - val_loss: 0.0052
Epoch 77/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - loss: 0.0078 - val_loss: 0.0052
Epoch 78/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step - loss: 0.0068 - val_loss: 0.0063
Epoch 79/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - loss: 0.0091 - val_loss: 0.0048
Epoch 80/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step - loss: 0.0093 - val_loss: 0.0049
Epoch 81/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0076 - val_loss: 0.0049
Epoch 82/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0082 - val_loss: 0.0051
Epoch 83/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0063 - val_loss: 0.0046
Epoch 84/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0074 - val_loss: 0.0045
Epoch 85/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0081 - val_loss: 0.0050
Epoch 86/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0090 - val_loss: 0.0045
Epoch 87/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0068 - val_loss: 0.0041
Epoch 88/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - loss: 0.0077 - val_loss: 0.0046
Epoch 89/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 14ms/step - loss: 0.0078 - val_loss: 0.0049
Epoch 90/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0067 - val_loss: 0.0042
Epoch 91/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0074 - val_loss: 0.0039
Epoch 92/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0070 - val_loss: 0.0038
Epoch 93/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0064 - val_loss: 0.0044
Epoch 94/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0073 - val_loss: 0.0037
Epoch 95/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - loss: 0.0067 - val_loss: 0.0037
Epoch 96/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0065 - val_loss: 0.0040
Epoch 97/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - loss: 0.0059 - val_loss: 0.0035
Epoch 98/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0066 - val_loss: 0.0037
Epoch 99/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0067 - val_loss: 0.0033
Epoch 100/100
10/10 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - loss: 0.0065 - val_loss: 0.0034
Model training finished.

Reasoning: Plot the training and validation loss from the model's training history to visualize the learning process.

# Plot the training and validation loss
plt.figure(figsize=(12, 6))
plt.plot(history.history['loss'], label='Training Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.title('Model Loss During Training')
plt.xlabel('Epoch')
plt.ylabel('Loss (MSE)')
plt.legend()
plt.show()

Realizar predicciones

Subtask:

Generar predicciones en el conjunto de prueba utilizando el modelo entrenado.

# Make predictions on the test set
predictions = model.predict(X_test)

print("Predictions on the test set have been generated.")

3/3 ━━━━━━━━━━━━━━━━━━━━ 0s 82ms/step
Predictions on the test set have been generated.

Post-procesamiento

Subtask:

Desnormalizar las predicciones y los valores reales del conjunto de prueba a la escala original de los datos.

# Inverse transform the predictions and actual values
predictions = scaler.inverse_transform(predictions)
y_test_actual = scaler.inverse_transform(y_test.reshape(-1, 1))

print("Predictions and actual values have been denormalized.")

Predictions and actual values have been denormalized.

Reasoning: Use the fitted scaler to inverse transform the scaled predictions and the actual y_test values to their original scale.

Reasoning: Use the trained model to predict the 'users' values on the testing set (X_test).

Evaluar el modelo

Subtask:

Calcular métricas de evaluación (MSE, RMSE, MAE, MAPE) para evaluar el rendimiento del modelo.

Reasoning: Import necessary metrics from sklearn.metrics and calculate Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) using the denormalized actual values and predictions.

from sklearn.metrics import mean_squared_error, mean_absolute_error
import numpy as np

# Calculate evaluation metrics
mse = mean_squared_error(y_test_actual, predictions)
rmse = np.sqrt(mse)
mae = mean_absolute_error(y_test_actual, predictions)
mape = np.mean(np.abs((y_test_actual - predictions) / y_test_actual)) * 100

print(f"Mean Squared Error (MSE): {mse:.2f}")
print(f"Root Mean Squared Error (RMSE): {rmse:.2f}")
print(f"Mean Absolute Error (MAE): {mae:.2f}")
print(f"Mean Absolute Percentage Error (MAPE): {mape:.2f}%")

Mean Squared Error (MSE): 150132.61
Root Mean Squared Error (RMSE): 387.47
Mean Absolute Error (MAE): 273.05
Mean Absolute Percentage Error (MAPE): 28.09%

Visualizar resultados

Subtask:

Generar los tres gráficos solicitados: (1) serie de tiempo completa mostrando train/test split con línea vertical divisoria, (2) zoom del período de prueba con valores observados vs predichos incluyendo fechas en el eje x, y (3) curvas de pérdida durante el entrenamiento (training y validation loss).

Reasoning: Plot the full time series, highlighting the training and testing periods with different colors and adding a vertical line to indicate the split point.

import matplotlib.pyplot as plt

# Get the dates for the test set
test_dates = df['date'][lookback_window + split_index:].reset_index(drop=True)

plt.figure(figsize=(15, 7))
plt.plot(df['date'], df['users'], label='Full Time Series')
plt.plot(df['date'][:lookback_window + split_index], df['users'][:lookback_window + split_index], label='Training Data', color='blue')
plt.plot(df['date'][lookback_window + split_index:], df['users'][lookback_window + split_index:], label='Testing Data', color='orange')
plt.plot(test_dates, predictions, label='Predicted Users', color='red', linestyle='--')

# Add a vertical line to show the split point
split_date = df['date'][lookback_window + split_index -1]
plt.axvline(split_date, color='red', linestyle='--', label='Train/Test Split')

plt.xlabel('Date')
plt.ylabel('Number of Users')
plt.title('Daily Website Users: Train and Test Split')
plt.legend()
plt.show()

Reasoning: Plot the denormalized actual values against the denormalized predictions for the test set, using the test dates on the x-axis for better context.

plt.figure(figsize=(15, 7))
plt.plot(test_dates, y_test_actual, label='Observed Users', color='blue')
plt.plot(test_dates, predictions, label='Predicted Users', color='orange', linestyle='--')

plt.xlabel('Date')
plt.ylabel('Number of Users')
plt.title('Observed vs Predicted Daily Website Users (Test Set)')
plt.legend()
plt.xticks(rotation=45)
plt.tight_layout()
plt.show()