alexanderchang1
commented
6 days ago Code to reproduce
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import torch # Import torch to handle tensors and CUDA devices
from contextualized.easy import ContextualizedBayesianNetworks
from sklearn.preprocessing import StandardScaler
# Detect CUDA device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Synthetic data example (replace this with your actual data)
n_samples = 50
n_genes = 10 # Number of genes
n_context_features = 20 # Number of context features (cna + rna)
# Generate random data for illustration purposes
np.random.seed(42)
C_array = np.random.randn(n_samples, n_context_features)
X_array = np.random.randn(n_samples, n_genes)
# Standardize the synthetic data
scaler_C = StandardScaler()
C_array = scaler_C.fit_transform(C_array)
scaler_X = StandardScaler()
X_array = scaler_X.fit_transform(X_array)
# Convert data to torch tensors and move them to the device
C_tensor = torch.tensor(C_array, dtype=torch.float32).to(device)
X_tensor = torch.tensor(X_array, dtype=torch.float32).to(device)
# Initialize the model (no need to move it to the device manually)
cbn = ContextualizedBayesianNetworks(
encoder_type='mlp',
num_archetypes=16,
n_bootstraps=2,
archetype_dag_loss_type="DAGMA",
archetype_alpha=0.,
sample_specific_dag_loss_type="DAGMA",
sample_specific_alpha=1e-1,
learning_rate=1e-3
)
# Train the model (the model expects tensors on the same device)
cbn.fit(C_tensor, X_tensor, max_epochs=100)
# Evaluate the model (again with data tensors on the same device)
mses = cbn.measure_mses(C_tensor, X_tensor)
print(f"Mean Squared Error: {np.mean(mses)}")
# Predict and visualize networks (ensure data stays on the correct device)
predicted_networks = cbn.predict_networks(C_tensor)
# Move the predicted network back to CPU for visualization
predicted_networks_cpu = predicted_networks[0].cpu().detach().numpy()
plt.imshow(predicted_networks_cpu)
plt.title("Predicted Network for First Sample")
plt.colorbar()
plt.show()
cnellington
Is there something I'm missing in the documentation as to how to move C-ML models and data to and from CUDA/CPU? I keep getting errors when trying to run code based on wrong location of torch.eye, etc. etc.