AmineTMIRI/backend
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add warm startu to model

Browse Source
This commit is contained in:
MuslemRahimi 2024-10-01 13:49:14 +02:00
parent 8872b5d1c7
commit d0b5cd5aaa
3 changed files with 115 additions and 193 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

183
app/cron_ai_score.py
View File

@ -38,23 +38,6 @@ def calculate_fdi(high, low, close, window=30):
return (2 - n1) * 100
def find_top_correlated_features(df, target_column, exclude_columns, top_n=10):
# Ensure the target column is not in the exclude list
exclude_columns = [col for col in exclude_columns if col != target_column]
# Select columns to consider for correlation
columns_to_consider = [col for col in df.columns if col not in exclude_columns + [target_column]]
# Calculate the correlation matrix
correlation_matrix = df[columns_to_consider + [target_column]].corr()
# Get correlations with the target column, excluding the target column itself
target_correlations = correlation_matrix[target_column].drop(target_column)
# Sort by absolute correlation value and select top N
top_correlated = target_correlations.abs().sort_values(ascending=False).head(top_n)
return top_correlated
async def download_data(ticker, con, start_date, end_date):
@ -63,14 +46,13 @@ async def download_data(ticker, con, start_date, end_date):
statements = [
f"json/financial-statements/ratios/quarter/{ticker}.json",
f"json/financial-statements/key-metrics/quarter/{ticker}.json",
#f"json/financial-statements/cash-flow-statement/quarter/{ticker}.json",
#f"json/financial-statements/income-statement/quarter/{ticker}.json",
#f"json/financial-statements/balance-sheet-statement/quarter/{ticker}.json",
f"json/financial-statements/cash-flow-statement/quarter/{ticker}.json",
f"json/financial-statements/income-statement/quarter/{ticker}.json",
f"json/financial-statements/balance-sheet-statement/quarter/{ticker}.json",
f"json/financial-statements/income-statement-growth/quarter/{ticker}.json",
f"json/financial-statements/balance-sheet-statement-growth/quarter/{ticker}.json",
f"json/financial-statements/cash-flow-statement-growth/quarter/{ticker}.json",
#f"json/financial-statements/key-metrics/quarter/{ticker}.json",
#f"json/financial-statements/owner-earnings/quarter/{ticker}.json",
f"json/financial-statements/owner-earnings/quarter/{ticker}.json",
]
# Helper function to load JSON data asynchronously
@ -91,35 +73,42 @@ async def download_data(ticker, con, start_date, end_date):
ratios = await load_json_from_file(statements[0])
ratios = await filter_data(ratios, ignore_keys)
#Threshold of enough datapoints needed!
if len(ratios) < 50:
return
key_metrics = await load_json_from_file(statements[1])
key_metrics = await filter_data(key_metrics, ignore_keys)
'''
cashflow = await load_json_from_file(statements[1])
cashflow = await load_json_from_file(statements[2])
cashflow = await filter_data(cashflow, ignore_keys)
income = await load_json_from_file(statements[2])
income = await load_json_from_file(statements[3])
income = await filter_data(income, ignore_keys)
balance = await load_json_from_file(statements[3])
balance = await load_json_from_file(statements[4])
balance = await filter_data(balance, ignore_keys)
'''
income_growth = await load_json_from_file(statements[2])
income_growth = await load_json_from_file(statements[5])
income_growth = await filter_data(income_growth, ignore_keys)
balance_growth = await load_json_from_file(statements[3])
balance_growth = await load_json_from_file(statements[6])
balance_growth = await filter_data(balance_growth, ignore_keys)
cashflow_growth = await load_json_from_file(statements[4])
cashflow_growth = await load_json_from_file(statements[7])
cashflow_growth = await filter_data(cashflow_growth, ignore_keys)
owner_earnings = await load_json_from_file(statements[8])
owner_earnings = await filter_data(owner_earnings, ignore_keys)
# Combine all the data
combined_data = defaultdict(dict)
# Merge the data based on 'date'
for entries in zip(ratios, key_metrics, income_growth, balance_growth, cashflow_growth):
for entries in zip(ratios, key_metrics, cashflow, income, balance, income_growth, balance_growth, cashflow_growth, owner_earnings):
for entry in entries:
date = entry['date']
for key, value in entry.items():
@ -321,29 +310,6 @@ async def download_data(ticker, con, start_date, end_date):
pass
async def process_symbol(ticker, con, start_date, end_date):
try:
test_size = 0.2
start_date = datetime(1995, 1, 1).strftime("%Y-%m-%d")
end_date = datetime.today().strftime("%Y-%m-%d")
predictor = ScorePredictor()
df = await download_data(ticker, con, start_date, end_date)
split_size = int(len(df) * (1-test_size))
test_data = df.iloc[split_size:]
selected_features = [col for col in df.columns if col not in ['date','price','Target']]
print(f"For the Ticker: {ticker}")
data = predictor.evaluate_model(test_data[selected_features], test_data['Target'])
if len(data) != 0:
if data['precision'] >= 50 and data['accuracy'] >= 50:
res = {'score': data['score']}
await save_json(ticker, res)
except Exception as e:
print(e)
async def chunked_gather(tickers, con, start_date, end_date, chunk_size=10):
# Helper function to divide the tickers into chunks
def chunks(lst, size):
@ -362,97 +328,84 @@ async def chunked_gather(tickers, con, start_date, end_date, chunk_size=10):
return results
#Train mode
async def train_process(tickers, con):
tickers = list(set(tickers))
df_train = pd.DataFrame()
df_test = pd.DataFrame()
test_size = 0.2
async def warm_start_training(tickers, con):
start_date = datetime(1995, 1, 1).strftime("%Y-%m-%d")
end_date = datetime.today().strftime("%Y-%m-%d")
df_train = pd.DataFrame()
df_test = pd.DataFrame()
dfs = await chunked_gather(tickers, con, start_date, end_date, chunk_size=10)
train_list = []
test_list = []
df_train = pd.concat(dfs, ignore_index=True)
df_train = df_train.sample(frac=1).reset_index(drop=True)
for df in dfs:
print('======Warm Start Train Set Datapoints======')
print(len(df_train))
predictor = ScorePredictor()
selected_features = [col for col in df_train if col not in ['price', 'date', 'Target']]
predictor.warm_start_training(df_train[selected_features], df_train['Target'])
return predictor
async def fine_tune_and_evaluate(ticker, con, start_date, end_date):
try:
df = await download_data(ticker, con, start_date, end_date)
if df is None or len(df) == 0:
print(f"No data available for {ticker}")
return
test_size = 0.2
split_size = int(len(df) * (1-test_size))
train_data = df.iloc[:split_size]
test_data = df.iloc[split_size:]
# Append to the lists
train_list.append(train_data)
test_list.append(test_data)
except:
pass
# Concatenate all at once outside the loop
df_train = pd.concat(train_list, ignore_index=True)
df_test = pd.concat(test_list, ignore_index=True)
best_features = [col for col in df_train.columns if col not in ['date','price','Target']]
df_train = df_train.sample(frac=1).reset_index(drop=True) #df_train.reset_index(drop=True)
top_correlated = find_top_correlated_features(df_train, 'Target', ['date', 'price'])
print(top_correlated)
#print(df_train)
print('======Train Set Datapoints======')
print(len(df_train))
selected_features = [col for col in df.columns if col not in ['date','price','Target']]
# Fine-tune the model
predictor = ScorePredictor()
#print(selected_features)
selected_features = [col for col in df_train if col not in ['price','date','Target']]
#best_features = predictor.feature_selection(df_train[selected_features], df_train['Target'],k=15)
#print(best_features)
predictor.train_model(df_train[selected_features], df_train['Target'])
predictor.evaluate_model(df_test[selected_features], df_test['Target'])
predictor.fine_tune_model(train_data[selected_features], train_data['Target'])
print(f"Evaluating fine-tuned model for {ticker}")
data = predictor.evaluate_model(test_data[selected_features], test_data['Target'])
if len(data) != 0:
if data['precision'] >= 50 and data['accuracy'] >= 50:
res = {'score': data['score']}
await save_json(ticker, res)
print(f"Saved results for {ticker}")
except Exception as e:
print(f"Error processing {ticker}: {e}")
async def run():
train_mode = True
train_mode = True # Set this to False for fine-tuning and evaluation
con = sqlite3.connect('stocks.db')
cursor = con.cursor()
cursor.execute("PRAGMA journal_mode = wal")
if train_mode:
#Train first model
cursor.execute("SELECT DISTINCT symbol FROM stocks WHERE marketCap >= 300E9 AND symbol NOT LIKE '%.%'")
stock_symbols = ['AAPL','AWR','TSLA','MSFT'] #[row[0] for row in cursor.fetchall()]
print('Number of Stocks')
print(len(stock_symbols))
await train_process(stock_symbols, con)
#Prediction Steps for all stock symbols
if not train_mode:
# Warm start training
warm_start_symbols = ['META', 'NFLX','GOOG','TSLA','AWR','AMD','NVDA']
print('Warm Start Training for:', warm_start_symbols)
predictor = await warm_start_training(warm_start_symbols, con)
else:
# Fine-tuning and evaluation for all stocks
cursor.execute("SELECT DISTINCT symbol FROM stocks WHERE marketCap >= 1E9 AND symbol NOT LIKE '%.%'")
stock_symbols = [row[0] for row in cursor.fetchall()]
total_symbols = stock_symbols
stock_symbols = ['NVDA'] #[row[0] for row in cursor.fetchall()]
print(f"Total tickers: {len(total_symbols)}")
print(f"Total tickers for fine-tuning: {len(stock_symbols)}")
start_date = datetime(1995, 1, 1).strftime("%Y-%m-%d")
end_date = datetime.today().strftime("%Y-%m-%d")
chunk_size = len(total_symbols) // 100 # Divide the list into N chunks
chunks = [total_symbols[i:i + chunk_size] for i in range(0, len(total_symbols), chunk_size)]
for chunk in chunks:
tasks = []
for ticker in tqdm(chunk):
tasks.append(process_symbol(ticker, con, start_date, end_date))
for ticker in tqdm(stock_symbols):
tasks.append(fine_tune_and_evaluate(ticker, con, start_date, end_date))
await asyncio.gather(*tasks)
con.close()
if __name__ == "__main__":
try:
asyncio.run(run())
except Exception as e:
print(e)
print(f"Main execution error: {e}")

BIN
app/ml_models/__pycache__/score_model.cpython-310.pyc
View File

Binary file not shown.

93
app/ml_models/score_model.py
View File

@ -56,96 +56,72 @@ class SelfAttention(Layer):
class ScorePredictor:
def __init__(self):
self.scaler = MinMaxScaler()
self.model = self.build_model()
self.model = None
self.warm_start_model_path = 'ml_models/weights/ai-score/warm_start_weights.keras'
def build_model(self):
clear_session()
# Input layer
inputs = Input(shape=(139,))
# First dense layer
x = Dense(128, activation='elu')(inputs)
inputs = Input(shape=(335,))
x = Dense(512, activation='elu')(inputs)
x = Dropout(0.2)(x)
x = BatchNormalization()(x)
# Additional dense layers
for units in [64, 32]:
x = Dense(units, activation='elu')(x)
x = Dropout(0.2)(x)
x = BatchNormalization()(x)
# Reshape for attention mechanism
x = Reshape((32, 1))(x)
# Attention mechanism
#attention = Dense(32, activation='elu')(x)
#attention = Dense(1, activation='softmax')(attention)
# Apply attention
#x = Multiply()([x, attention])
x, _ = SelfAttention()(x)
outputs = Dense(2, activation='softmax')(x)
# Global average pooling
#x = GlobalAveragePooling1D()(x)
# Output layer (for class probabilities)
outputs = Dense(2, activation='softmax')(x) # Two neurons for class probabilities with softmax
# Create the model
model = Model(inputs=inputs, outputs=outputs)
# Optimizer with a lower learning rate
optimizer = Adam(learning_rate=0.01, clipnorm=1.0)
# Compile the model
model.compile(optimizer=optimizer, loss='sparse_categorical_crossentropy', metrics=['accuracy'])
return model
def preprocess_data(self, X):
# X = X.applymap(lambda x: 9999 if x == 0 else x) # Replace 0 with 9999 as suggested in the paper
X = np.where(np.isinf(X), np.nan, X)
X = np.nan_to_num(X)
X = self.scaler.fit_transform(X)
return X
def reshape_for_lstm(self, X):
return X.reshape((X.shape[0], X.shape[1], 1))
def train_model(self, X_train, y_train):
def warm_start_training(self, X_train, y_train):
X_train = self.preprocess_data(X_train)
#X_train = self.reshape_for_lstm(X_train)
self.model = self.build_model()
checkpoint = ModelCheckpoint('ml_models/weights/ai-score/weights.keras',
save_best_only=True, save_freq = 1,
monitor='val_loss', mode='min')
checkpoint = ModelCheckpoint(self.warm_start_model_path, save_best_only=True, save_freq=1, monitor='val_loss', mode='min')
early_stopping = EarlyStopping(monitor='val_loss', patience=20, restore_best_weights=True)
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=10, min_lr=0.001)
self.model.fit(X_train, y_train, epochs=100_000, batch_size=32,
validation_split=0.1, callbacks=[checkpoint, early_stopping, reduce_lr])
self.model.save('ml_models/weights/ai-score/weights.keras')
self.model.fit(X_train, y_train, epochs=100_000, batch_size=32, validation_split=0.1, callbacks=[checkpoint, early_stopping, reduce_lr])
self.model.save(self.warm_start_model_path)
print("Warm start model saved.")
def fine_tune_model(self, X_train, y_train):
X_train = self.preprocess_data(X_train)
if self.model is None:
self.model = load_model(self.warm_start_model_path, custom_objects={'SelfAttention': SelfAttention})
early_stopping = EarlyStopping(monitor='val_loss', patience=20, restore_best_weights=True)
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=10, min_lr=0.0001)
self.model.fit(X_train, y_train, epochs=100, batch_size=32, validation_split=0.1, callbacks=[early_stopping, reduce_lr])
print("Model fine-tuned (not saved).")
def evaluate_model(self, X_test, y_test):
# Preprocess the test data
X_test = self.preprocess_data(X_test)
#X_test = self.reshape_for_lstm(X_test)
# Load the trained model
self.model = load_model('ml_models/weights/ai-score/weights.keras')
if self.model is None:
raise ValueError("Model has not been trained or fine-tuned. Call warm_start_training or fine_tune_model first.")
# Get the model's predictions
test_predictions = self.model.predict(X_test)
print(test_predictions)
# Extract the probabilities for class 1 (index 1 in the softmax output)
class_1_probabilities = test_predictions[:, 1]
# Convert probabilities to binary predictions using a threshold of 0.5
binary_predictions = (class_1_probabilities >= 0.5).astype(int)
# Calculate precision and accuracy using binary predictions
print(test_predictions)
test_precision = precision_score(y_test, binary_predictions)
test_accuracy = accuracy_score(y_test, binary_predictions)
@ -153,32 +129,25 @@ class ScorePredictor:
print(f"Precision: {round(test_precision * 100)}%")
print(f"Accuracy: {round(test_accuracy * 100)}%")
# Define thresholds and corresponding scores
thresholds = [0.8, 0.75, 0.7, 0.6, 0.5, 0.45, 0.4, 0.35, 0.3, 0.2]
scores = [10, 9, 8, 7, 6, 5, 4, 3, 2, 1]
# Get the last prediction value (class 1 probability) for scoring
last_prediction_prob = class_1_probabilities[-1]
# Initialize score to 0 (or any default value)
score = 0
print(last_prediction_prob)
# Determine the score based on the last prediction probability
print(f"Last prediction probability: {last_prediction_prob}")
for threshold, value in zip(thresholds, scores):
if last_prediction_prob >= threshold:
score = value
break # Exit the loop once the score is determined
break
# Return the evaluation results
return {'accuracy': round(test_accuracy * 100),
'precision': round(test_precision * 100),
'score': score}
def feature_selection(self, X_train, y_train, k=100):
print('feature selection:')
print(X_train.shape, y_train.shape)
print('Feature selection:')
print(f"X_train shape: {X_train.shape}, y_train shape: {y_train.shape}")
selector = SelectKBest(score_func=f_classif, k=k)
selector.fit(X_train, y_train)