Skip to content
This repository has been archived by the owner on May 18, 2024. It is now read-only.

Commit

Permalink
Some Changes! Works better now!
Browse files Browse the repository at this point in the history
  • Loading branch information
@lex-hue
lex-hue committed Jun 17, 2023
1 parent f06fe5a commit a98abd6
Show file tree
Hide file tree
Showing 3 changed files with 118 additions and 94 deletions.
3 changes: 2 additions & 1 deletion README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ deactivate # Execute this command to deactivate the virtual environment
### 1c. Updating the Repository
```
python SPV4.py --update
python SPV4.py --install
```

---
Expand Down Expand Up @@ -93,7 +94,7 @@ python SPV4.py --fine_tune
```
Before starting the fine-tuning process for the LSTM RL model, the script will now ask you to specify your desired reward threshold. It is recommended to set the threshold at 0.9.

While waiting for the fine-tuning process to finish, it is advisable to stay hydrated by drinking some water, as this process may take some time. Once the fine-tuning is complete, try running the script until it loops only once. In case it doesn't loop once initially, continue rerunning it until it completes at least one loop.
It is recommended to maintain hydration by drinking water while waiting for the fine-tuning process to complete, as this may take a while. Once the fine-tuning is done, attempt to run the script with a single loop. If the script doesn't loop once initially, keep rerunning it until it completes at least one loop. If the script consistently produces the same result without any improvement, please interrupt it by pressing CTRL + C, answer "yes" to the prompt, and rerun the script.

After the fine-tuning process is fully complete, it is highly recommended to re-evaluate the model's performance.

Expand Down
207 changes: 115 additions & 92 deletions SPV4.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -77,7 +77,7 @@ def install_dependencies():
"tensorflow-cpu",
"matplotlib",
"ta-lib",
"bayesian-optimization",
"optuna",
]
total_packages = len(packages)
progress = 0
Expand Down Expand Up @@ -283,7 +283,8 @@ def train_model():
)
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
from sklearn.metrics import mean_absolute_percentage_error, r2_score
from bayes_opt import BayesianOptimization
from optuna import create_study, Trial, visualization
from optuna.samplers import TPESampler

print("Training the SPV4 model...")
print("TensorFlow version:", tf.__version__)
Expand Down Expand Up @@ -344,59 +345,100 @@ def create_sequences(data, timesteps):
X_test, y_test = create_sequences(test_data_norm, timesteps)

# Define the Deep RL model
def create_model(units, filters, kernel_size, learning_rate):
def create_model(trial):
model = Sequential()
model.add(Conv1D(filters=filters, kernel_size=kernel_size, activation="relu"))
model.add(
Conv1D(
filters=trial.suggest_int("filters", 50, 450),
kernel_size=trial.suggest_int("kernel_size", 2, 15),
activation="relu"
)
)
model.add(MaxPooling1D(pool_size=2))
model.add(Conv1D(filters=filters // 2, kernel_size=kernel_size // 2, activation="relu"))
model.add(LSTM(units=units, return_sequences=True, input_shape=(timesteps, X_train.shape[2])))
model.add(
Conv1D(
filters=trial.suggest_int("filters_2", 50, 450),
kernel_size=trial.suggest_int("kernel_size_2", 1, 15),
activation="relu"
)
)
model.add(
LSTM(
units=trial.suggest_int("units", 50, 300),
return_sequences=True,
input_shape=(timesteps, X_train.shape[2])
)
)
model.add(BatchNormalization())
model.add(LSTM(units=units, return_sequences=True))
model.add(
LSTM(
units=trial.suggest_int("units_2", 50, 300),
return_sequences=True
)
)
model.add(BatchNormalization())
model.add(Dense(units=units))
model.add(Dense(units=trial.suggest_int("units_3", 50, 300)))
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(units=units)))
model.add(TimeDistributed(Dense(units=trial.suggest_int("units_4", 50, 300))))
model.add(BatchNormalization())
model.add(LSTM(units=units // 2, return_sequences=True))
model.add(
LSTM(
units=trial.suggest_int("units_5", 25, 150),
return_sequences=True
)
)
model.add(BatchNormalization())
model.add(LSTM(units=units // 2, return_sequences=True))
model.add(
LSTM(
units=trial.suggest_int("units_6", 25, 150),
return_sequences=True
)
)
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(units=units // 2)))
model.add(TimeDistributed(Dense(units=trial.suggest_int("units_7", 25, 150))))
model.add(BatchNormalization())
model.add(LSTM(units=units // 4))
model.add(LSTM(units=trial.suggest_int("units_8", 12, 75)))
model.add(BatchNormalization())
model.add(Dense(units=1))

# Define the RL optimizer and compile the model
optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)
optimizer = tf.keras.optimizers.Adam(learning_rate=trial.suggest_float("learning_rate", 0.001, 0.015))
model.compile(optimizer=optimizer, loss="mse")

return model

# Define RL training loop
epochs = 20
epochs = 10
epochs1 = 3
batch_size = 50
best_reward = None
min_reward_threshold = float(input("Enter the minimum reward threshold (e.g., 0.7): "))

def optimize_model(units, filters, kernel_size, learning_rate):
model = create_model(int(units), int(filters), int(kernel_size), learning_rate)
def optimize_model(trial: Trial):
model = create_model(trial)

for i in range(epochs1):
print("Epoch", i, "/", epochs1)
# Train the model for one epoch
for a in range(0, len(X_train), batch_size):
if a == 0:
print("Batch", a, "/", len(X_train), "(", ((a/len(X_train))*100), "% Done)")
else:
sys.stdout.write('\033[F\033[K')
print("Batch", a, "/", len(X_train), "(", ((a/len(X_train))*100), "% Done)")
batch_X = X_train[a:a + batch_size]
batch_y = y_train[a:a + batch_size]
history = model.fit(batch_X, batch_y, batch_size=batch_size, epochs=1, verbose=0)
sys.stdout.write('\033[F\033[K')
sys.stdout.write('\033[F\033[K')
print("Epoch", i+1, "/", epochs1)
# Train the model for one epoch
for a in range(0, len(X_train), batch_size):
if a == 0:
print(
"Batch", a+1, "/", len(X_train),
"(", ((a/len(X_train))*100), "% Done)"
)
else:
sys.stdout.write('\033[F\033[K')
print(
"Batch", a+1, "/", len(X_train),
"(", ((a/len(X_train))*100), "% Done)"
)
batch_X = X_train[a:a + batch_size]
batch_y = y_train[a:a + batch_size]
history = model.fit(
batch_X, batch_y,
batch_size=batch_size, epochs=1, verbose=0
)
sys.stdout.write('\033[F\033[K')
sys.stdout.write('\033[F\033[K')

# Evaluate the model on the test set
y_pred_test = model.predict(X_test)
Expand All @@ -405,76 +447,58 @@ def optimize_model(units, filters, kernel_size, learning_rate):

return test_reward

# Define the search space boundaries
pbounds = {
'units': (50, 300),
'filters': (50, 450),
'kernel_size': (2, 15),
'learning_rate': (0.001, 0.015),
}
# Define the search space boundaries and create Optuna study
sampler = TPESampler(seed=42)
study = create_study(sampler=sampler, direction="maximize")
study.optimize(optimize_model, n_trials=5)

# Define Bayesian optimization function
optimizer = BayesianOptimization(f=optimize_model, pbounds=pbounds)

# Perform Bayesian optimization
num_iterations = 2

for a in range(num_iterations):
print(((a/num_iterations)*100), "% Done")
optimizer.maximize(init_points=2, n_iter=2)
best_params = optimizer.max['params']
best_reward = optimizer.max['target']

# Check if the minimum reward threshold is reached
if best_reward >= min_reward_threshold:
break
# Get best parameters and reward
best_trial = study.best_trial
best_params = best_trial.params
best_reward = best_trial.value

print("\nBest reward:", best_reward)
print("Best parameters:", best_params)

best_reward1 = None
model_saved = 0

# Load the best model and evaluate it
print("Training best model")
while model_saved == 0:
model = create_model(int(best_params['units']), int(best_params['filters']), int(best_params['kernel_size']), best_params['learning_rate'])
for i in range(epochs):
print("Epoch", i, "/", epochs)
# Train the model for one epoch
for a in range(0, len(X_train), batch_size):
if a == 0:
print("Batch", a, "/", len(X_train), "(", ((a/len(X_train))*100), "% Done)")
else:
sys.stdout.write('\033[F\033[K')
print("Batch", a, "/", len(X_train), "(", ((a/len(X_train))*100), "% Done)")
batch_X = X_train[a:a + batch_size]
batch_y = y_train[a:a + batch_size]
history = model.fit(batch_X, batch_y, batch_size=batch_size, epochs=1, verbose=0)
model = create_model(best_trial)
for i in range(epochs):
print("Epoch", i+1, "/", epochs)
# Train the model for one epoch
for a in range(0, len(X_train), batch_size):
if a == 0:
print(
"Batch", a+1, "/", len(X_train),
"(", ((a/len(X_train))*100), "% Done)"
)
else:
sys.stdout.write('\033[F\033[K')
print(
"Batch", a+1, "/", len(X_train),
"(", ((a/len(X_train))*100), "% Done)"
)
batch_X = X_train[a:a + batch_size]
batch_y = y_train[a:a + batch_size]
history = model.fit(
batch_X, batch_y,
batch_size=batch_size, epochs=1, verbose=0
)

# Evaluate the model on the test set
y_pred_test = model.predict(X_test)
sys.stdout.write('\033[F\033[K')
test_reward = get_reward(y_test, y_pred_test)
# Evaluate the model on the test set
y_pred_test = model.predict(X_test)
sys.stdout.write('\033[F\033[K')
test_reward = get_reward(y_test, y_pred_test)

print("Test reward:", test_reward)
print("Test reward:", test_reward)

if i == 0:
best_reward1 = test_reward
if i == 0:
best_reward1 = test_reward

if test_reward >= best_reward1 and test_reward >= best_reward:
print("Model saved!")
model_saved = 1
best_reward1 = test_reward
model.save("model.h5")

if test_reward >= min_reward_threshold:
print("Model reached reward threshold", test_reward, ". Saving and stopping epochs!")
model_saved = 1
model.save("model.h5")
break
if test_reward >= best_reward1:
print("Model saved!")
model.save("model.h5")

else:
if i == epochs - 1:
model = load_model("model.h5")
y_pred_test = model.predict(X_test)
test_reward = get_reward(y_test, y_pred_test)
Expand Down Expand Up @@ -707,11 +731,10 @@ def handle_interrupt(signal, frame):

# Ask the user for confirmation
user_input = input(
f"Are you sure that you want to save the Model and also End the Program? (yes/no): "
f"Are you sure that you want to End the Program? (yes/no): "
)

if user_input.lower() == "yes":
model.save("model.h5")
exit(0)

else:
Expand Down
2 changes: 1 addition & 1 deletion commit_sha.sha
View file
Original file line number Diff line number Diff line change
@@ -1,2 +1,2 @@
468168b219440fa94def318347b53149cd13a8c1c7c1c5a36b6fa747e57c04a8
30d27ed61dc3d4358f60947bde457ece7f90d3a30ef98803d22910440aceccc4
Major Update

0 comments on commit a98abd6

Please sign in to comment.