基于Keras实现多变量预测：利用情感值预测价格

Hey there! Let's walk through how to build your Price prediction model with Keras using your sentiment-dependent dataset.

Dataset Overview #

You’ve got a 3-column dataset where Price movement is directly tied to Sentiment (positive sentiment pushes prices up, negative sentiment pulls them down):

Step 1: Data Preprocessing #

First, we need to get the data ready for Keras. Here’s what to tackle:

• Convert categorical Sentiment values to numerical format (e.g., map "positive" to 1, "negative" to -1, "neutral" to 0).
• Split your data into training and test sets (standard 80-20 split works well).
• Reshape inputs to fit Keras’ expected 2D array format: (number_of_samples, number_of_features).

Sample preprocessing code:

import pandas as pd
from sklearn.model_selection import train_test_split

# Load your dataset
df = pd.read_csv("your_dataset.csv")

# Encode sentiment (adjust mapping to match your actual sentiment labels)
sentiment_mapping = {"positive": 1, "negative": -1, "neutral": 0}
df["Sentiment"] = df["Sentiment"].map(sentiment_mapping)

# Define input features (X) and target variable (y)
X = df[["Sentiment"]].values
y = df["Price"].values

# Split into training and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

Step 2: Build the Keras Model #

Since Price depends directly on Sentiment, a simple feedforward neural network is a great starting point. We’ll use dense layers to capture the relationship:

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

# Initialize the sequential model
model = Sequential()

# Input layer + first hidden layer (1 input feature = Sentiment)
model.add(Dense(16, activation='relu', input_shape=(1,)))
# Second hidden layer to capture complex patterns
model.add(Dense(8, activation='relu'))
# Output layer (predicting continuous Price, so no activation function)
model.add(Dense(1))

# Compile the model: use Adam optimizer and MSE loss for regression tasks
model.compile(optimizer='adam', loss='mean_squared_error')

Step 3: Train the Model #

Now let’s train the model on our preprocessed data:

# Train the model with validation split to monitor overfitting
history = model.fit(
    X_train, y_train,
    epochs=50,
    batch_size=4,
    validation_split=0.1
)

Step 4: Evaluate and Predict #

After training, check performance on the test set and generate predictions:

# Evaluate model on test data
test_loss = model.evaluate(X_test, y_test)
print(f"Test Set MSE Loss: {test_loss}")

# Generate price predictions for test data
predictions = model.predict(X_test)

Quick Tips to Boost Performance #

• If your Sentiment values are already continuous (not categorical), skip the encoding step and use raw values directly.
• Add temporal features from the Date column (e.g., day of week, month) if price has time-based patterns.
• Adjust the number of neurons/layers or try different activation functions if the model is underfitting/overfitting.

内容的提问来源于stack exchange，提问作者Nidhin Radh