IMDB / train.py
4loop's picture
Initial commit: Clean IMDB Sentiment Analysis project
783f1c5
Raw History Blame Contribute Delete
3.92 kB
import os
import json
import mlflow
import mlflow.keras
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.datasets import imdb
from tensorflow.keras.preprocessing.sequence import pad_sequences
# ---------------------------
# 1. Hyperparameters
# ---------------------------
vocab_size = 10000
maxlen = 200
embedding_dim = 16
epochs = 3 # Reduced for faster training
batch_size = 64
# ---------------------------
# 2. Load and preprocess data
# ---------------------------
print("Loading IMDB dataset...")
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=vocab_size)
print("Padding sequences...")
x_train = pad_sequences(x_train, maxlen=maxlen)
x_test = pad_sequences(x_test, maxlen=maxlen)
print("Preparing word index...")
# Get and adjust word index (IMDB specific)
word_index = imdb.get_word_index()
word_index = {k:(v+3) for k,v in word_index.items()}
word_index["<PAD>"] = 0
word_index["<START>"] = 1
word_index["<UNK>"] = 2 # Out-of-vocabulary
word_index["<UNUSED>"] = 3
# Reverse word index for decoding
reverse_word_index = {v: k for k, v in word_index.items()}
# ---------------------------
# 3. Build model
# ---------------------------
print("Building model...")
model = keras.Sequential([
keras.layers.Embedding(vocab_size, embedding_dim, input_length=maxlen),
keras.layers.GlobalAveragePooling1D(), # Better than Flatten for sequences
keras.layers.Dense(16, activation="relu"),
keras.layers.Dropout(0.5),
keras.layers.Dense(1, activation="sigmoid")
])
model.compile(
optimizer="adam",
loss="binary_crossentropy",
metrics=["accuracy"]
)
print(model.summary())
# ---------------------------
# 4. Train + MLflow logging
# ---------------------------
print("Starting training with MLflow logging...")
# Set MLflow tracking (local by default)
mlflow.set_tracking_uri("file:./mlruns") # Local storage
mlflow.set_experiment("imdb-sentiment")
with mlflow.start_run() as run:
run_id = run.info.run_id
print(f"✅ Training started. Run ID: {run_id}")
# Log parameters
mlflow.log_params({
"vocab_size": vocab_size,
"embedding_dim": embedding_dim,
"maxlen": maxlen,
"epochs": epochs,
"batch_size": batch_size,
"model_type": "Embedding+GlobalAveragePooling"
})
# Train model
history = model.fit(
x_train, y_train,
epochs=epochs,
batch_size=batch_size,
validation_split=0.2,
verbose=1
)
# Evaluate
test_loss, test_accuracy = model.evaluate(x_test, y_test, verbose=0)
print(f"✅ Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.4f}")
# Log metrics
mlflow.log_metrics({
"test_loss": test_loss,
"test_accuracy": test_accuracy
})
# Log final training metrics
final_train_acc = history.history['accuracy'][-1]
final_val_acc = history.history['val_accuracy'][-1]
mlflow.log_metrics({
"final_train_accuracy": final_train_acc,
"final_val_accuracy": final_val_acc
})
# Log model
mlflow.keras.log_model(
model,
"model",
registered_model_name="IMDBSentimentModel"
)
# ---------------------------
# 5. Save artifacts for deployment
# ---------------------------
os.makedirs("model", exist_ok=True)
# Save model in Keras format
model.save("model/saved_model.keras")
# Save word index
with open("model/word_index.json", "w") as f:
json.dump(word_index, f)
# Save reverse word index for text generation
with open("model/reverse_word_index.json", "w") as f:
json.dump(reverse_word_index, f)
# Log artifacts
mlflow.log_artifacts("model", artifact_path="model")
print("✅ Training complete!")
print("✅ Model saved locally in './model/' directory")
print("✅ Model registered with MLflow as 'IMDBSentimentModel'")