Merge pull request #42 from redBorder/add_default_model_logic

PR-39: Add default model logic to the api

resources/src/ai/outliers.py

@@ -26,12 +26,16 @@
 '''
 End of important OS Variables
 '''
+import sys
 import shutil
 import numpy as np
 import configparser
 import pandas as pd
 import tensorflow as tf
 
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
+from logger import logger
+
 class Autoencoder:
     """
     Autoencoder class for anomaly detection.
@@ -47,14 +51,14 @@ def __init__(self, model_file, model_config_file):
         Args:
             model_file (str): Path to model .keras file.
             model_config_file (str): Path to the model config, including:
-                METRICS (list): Names of the metrics used by the module.
-                TIMESTAMP (list): Names of the timestamp columns used by the module.
-                AVG_LOSS (float): Average loss of the model.
-                STD_LOSS (float): Standard deviation of the loss of the model.
-                WINDOW_SIZE (int): Number of entries the model will put together in a 'window'.
-                NUM_WINDOWS (int): Number of windows the model will put together in each slice.
-                LOSS_MULT_1 (float): Extra penalty in the loss function for guessing wrong metrics.
-                LOSS_MULT_2 (float): Extra penalty in the loss function for guessing wrong 
+                metrics (list): Names of the metrics used by the module.
+                timestamp (list): Names of the timestamp columns used by the module.
+                avg_loss (float): Average loss of the model.
+                std_loss (float): Standard deviation of the loss of the model.
+                window_size (int): Number of entries the model will put together in a 'window'.
+                num_window (int): Number of windows the model will put together in each slice.
+                loss_mult_metric (float): Extra penalty in the loss function for guessing wrong metrics.
+                loss_mult_minute (float): Extra penalty in the loss function for guessing wrong
                   'minute' field.
         """
         self.check_existence(model_file, model_config_file)
@@ -63,36 +67,36 @@ def __init__(self, model_file, model_config_file):
             model_config = configparser.ConfigParser()
             model_config.read(model_config_file)
             columns_section = model_config['Columns']
-            self.METRICS = columns_section.get('METRICS', '').split(', ')
-            self.TIMESTAMP = columns_section.get('TIMESTAMP', '').split(', ')
-            self.COLUMNS = self.METRICS + self.TIMESTAMP
+            self.metrics = columns_section.get('METRICS', '').split(', ')
+            self.timestamp = columns_section.get('TIMESTAMP', '').split(', ')
+            self.columns = self.metrics + self.timestamp
             general_section = model_config['General']
-            self.AVG_LOSS = float(general_section.get('AVG_LOSS', 0.0))
-            self.STD_LOSS = float(general_section.get('STD_LOSS', 0.0))
-            self.WINDOW_SIZE = int(general_section.get('WINDOW_SIZE', 0))
-            self.NUM_WINDOWS = int(general_section.get('NUM_WINDOWS', 0))
-            self.LOSS_MULT_1 = float(general_section.get('LOSS_MULT_1', 0))
-            self.LOSS_MULT_2 = float(general_section.get('LOSS_MULT_2', 0))
+            self.avg_loss = float(general_section.get('AVG_LOSS', 0.0))
+            self.std_loss = float(general_section.get('STD_LOSS', 0.0))
+            self.window_size = int(general_section.get('WINDOW_SIZE', 0))
+            self.num_window = int(general_section.get('NUM_WINDOWS', 0))
+            self.loss_mult_metric = float(general_section.get('LOSS_MULT_METRIC', 0))
+            self.loss_mult_minute = float(general_section.get('LOSS_MULT_MINUTE', 0))
         except FileNotFoundError:
-            print(f"Error: Model file '{model_config_file}' not found.")
+            logger.logger.error(f"Error: Model file '{model_config_file}' not found.")
         except (OSError, ValueError) as e:
-            print(f"Error loading model conif: {e}")
+            logger.logger.error(f"Error loading model conif: {e}")
         try:
             self.model = tf.keras.models.load_model(
                 model_file,
                 compile=False
             )
         except FileNotFoundError:
-            print(f"Error: Model file '{model_file}' not found.")
+            logger.logger.error(f"Error: Model file '{model_file}' not found.")
         except (OSError, ValueError) as e:
-            print(f"Error loading the model: {e}")
+            logger.logger.error(f"Error loading the model: {e}")
 
     def check_existence(self, model_file, model_config_file):
         """
         Check existence of model files and copy them if missing.
 
-        This function checks if the provided `model_file` and `model_config_file` exist in their 
-        respective paths. If they don't exist, it renames and copies the corresponding default 
+        This function checks if the provided `model_file` and `model_config_file` exist in their
+        respective paths. If they don't exist, it renames and copies the corresponding default
         files from the 'traffic.keras' and 'traffic.ini' files, which are expected to be located
         in the same directory as the target files.
 
@@ -122,7 +126,7 @@ def rescale(self, data):
         Returns:
             (numpy.ndarray): Rescaled data as a numpy array.
         """
-        num_metrics = len(self.METRICS)
+        num_metrics = len(self.metrics)
         rescaled=data.copy()
         rescaled[..., 0:num_metrics]=np.tanh(np.log1p(rescaled[..., 0:num_metrics])/32)
         rescaled[..., num_metrics]=rescaled[..., num_metrics]/1440
@@ -138,7 +142,7 @@ def descale(self, data):
         Returns:
             (numpy.ndarray): Descaled data as a numpy array.
         """
-        num_metrics = len(self.METRICS)
+        num_metrics = len(self.metrics)
         descaled = data.copy()
         descaled = np.where(descaled > 1.0, 1.0, np.where(descaled < -1.0, -1.0, descaled))
         descaled[..., 0:num_metrics] = np.expm1(32*np.arctanh(descaled[..., 0:num_metrics]))
@@ -154,7 +158,7 @@ def model_loss(self, y_true, y_pred, single_value=True):
         otherwise, the value is left unchanged.
         Then, the difference between both tensors is evaluated and a log_cosh loss
         is applied.
-        
+
         Args:
             y_true (tf.Tensor): True target values.
             y_pred (tf.Tensor): Predicted values.
@@ -165,25 +169,24 @@ def model_loss(self, y_true, y_pred, single_value=True):
         """
         y_true = tf.cast(y_true, tf.float16)
         y_pred = tf.cast(y_pred, tf.float16)
-        num_metrics = len(self.METRICS)
-        num_features = len(self.COLUMNS)
+        num_metrics = len(self.metrics)
+        num_features = len(self.columns)
         is_metric = (tf.range(num_features) < num_metrics)
         is_minute = (tf.range(num_features) == num_metrics)
         mult_true = tf.where(
-            is_metric, self.LOSS_MULT_1 * y_true,
-            tf.where(is_minute, self.LOSS_MULT_2 * y_true, y_true)
+            is_metric, self.loss_mult_metric * y_true,
+            tf.where(is_minute, self.loss_mult_minute * y_true, y_true)
         )
         mult_pred = tf.where(
-            is_metric, self.LOSS_MULT_1 * y_pred,
-            tf.where(is_minute, self.LOSS_MULT_2 * y_pred, y_pred)
+            is_metric, self.loss_mult_metric * y_pred,
+            tf.where(is_minute, self.loss_mult_minute * y_pred, y_pred)
         )
         standard_loss = tf.math.log(tf.cosh((mult_true - mult_pred)))
         if single_value:
             standard_loss = tf.reduce_mean(standard_loss)
         return standard_loss
 
     def slice(self, data, index = []):
-        #TODO add a graph to doc to explain this
         """
         Transform a 2D numpy array into a 3D array readable by the model.
 
@@ -195,13 +198,13 @@ def slice(self, data, index = []):
             (numpy.ndarray): 3D numpy array that can be processed by the model.
         """
         _l = len(data)
-        Xs = []
-        slice_length = self.WINDOW_SIZE * self.NUM_WINDOWS
+        sliced_data = []
+        slice_length = self.window_size * self.num_window
         if len(index) == 0:
-            index = np.arange(0, _l-slice_length+1 , self.WINDOW_SIZE)
+            index = np.arange(0, _l-slice_length+1 , self.window_size)
         for i in index:
-            Xs.append(data[i:i+slice_length])
-        return np.array(Xs)
+            sliced_data.append(data[i:i+slice_length])
+        return np.array(sliced_data)
 
     def flatten(self, data):
         """
@@ -213,11 +216,11 @@ def flatten(self, data):
         """
         tsr = data.copy()
         num_slices, slice_len, features = tsr.shape
-        flattened_len = (num_slices-1)*self.WINDOW_SIZE + slice_len
+        flattened_len = (num_slices-1)*self.window_size + slice_len
         flattened_tensor = np.zeros([flattened_len, features])
         scaling = np.zeros(flattened_len)
         for i in range(num_slices):
-            left_pad = i*self.WINDOW_SIZE
+            left_pad = i*self.window_size
             right_pad = left_pad+slice_len
             flattened_tensor[left_pad:right_pad] += tsr[i]
             scaling[left_pad:right_pad] +=1
@@ -256,10 +259,10 @@ def compute_json(self, metric, raw_json):
             (Json): Json with the anomalies and predictions for the data with RedBorder
               prediction Json format.
         """
-        threshold = self.AVG_LOSS+5*self.STD_LOSS
+        threshold = self.avg_loss+5*self.std_loss
         data, timestamps = self.input_json(raw_json)
         predicted, loss = self.calculate_predictions(data)
-        predicted = pd.DataFrame(predicted, columns=self.COLUMNS)
+        predicted = pd.DataFrame(predicted, columns=self.columns)
         predicted['timestamp'] = timestamps
         anomalies = predicted[loss>threshold]
         return self.output_json(metric, anomalies, predicted)
@@ -271,7 +274,7 @@ def granularity_from_dataframe(self, dataframe):
 
         Args:
             dataframe (pandas.DataFrame): Dataframe with timestamp column
-        
+
         Returns:
             time_diffs (pandas.Series): Series with the estimated Granularity of the dataframe.
         """
@@ -287,22 +290,23 @@ def input_json(self, raw_json):
 
         Args:
             raw_json (Json): druid Json response with the data.
-        
+
         Returns:
             data (numpy.ndarray): transformed data.
-            timestamps (pandas.Series): pandas series with the timestamp of each entry. 
+            timestamps (pandas.Series): pandas series with the timestamp of each entry.
         """
         data = pd.json_normalize(raw_json)
         data["granularity"] = self.granularity_from_dataframe(data)
-        metrics_dict = {f"result.{metric}": metric for metric in self.METRICS}
+        metrics_dict = {f"result.{metric}": metric for metric in self.metrics}
         data.rename(columns=metrics_dict, inplace=True)
         timestamps = data['timestamp'].copy()
         data['timestamp'] = pd.to_datetime(data['timestamp'])
         data['minute'] = data['timestamp'].dt.minute + 60 * data['timestamp'].dt.hour
-        data = pd.get_dummies(data, columns=['timestamp'], prefix=['weekday'], drop_first=True)
-        missing_columns = set(self.COLUMNS) - set(data.columns)
+        data['weekday']= data['timestamp'].dt.weekday
+        data = pd.get_dummies(data, columns=['weekday'], prefix=['weekday'], drop_first=True)
+        missing_columns = set(self.columns) - set(data.columns)
         data[list(missing_columns)] = 0
-        data = data[self.COLUMNS].dropna()
+        data = data[self.columns].dropna().astype('float')
         data_array = data.values
         return data_array, timestamps
 

resources/src/ai/shallow_outliers.py

@@ -38,7 +38,7 @@ def predict(self, arr):
 
         Args:
             arr (numpy.ndarray): 1D numpy array with the datapoints to be smoothed.
-        
+
         Returns:
             smooth_arr (numpy.ndarray): 1D numpy array with the smoothed data. Same shape as arr.
         """
@@ -66,15 +66,15 @@ def get_outliers(self, arr, smoothed_arr):
         an outlier and False otherwise.
 
         The method used for outlier detection is an isolation forest, which will look for
-        the 0.3% most isolated points when taking into account the original value, the 
+        the 0.3% most isolated points when taking into account the original value, the
         smoothed valued, the diference between them (error) and the squared diference
         between them.
 
         Args:
             arr (numpy.ndarray): 1D numpy array where the outliers shall be detected.
             smoothed_arr (numpy.ndarray): 1D numpy array that tries to approximate arr.
                 -Must have the same shape as arr.
-        
+
         Returns:
             numpy.ndarray: 1D numpy array with the smoothed data.
         """

resources/src/ai/traffic.ini

@@ -7,6 +7,6 @@ avg_loss = 0.09741491896919627
 std_loss = 0.11098885675977664
 window_size = 16
 num_windows = 2
-loss_mult_1 = 20.0
-loss_mult_2 = 10.0
+loss_mult_metric = 20.0
+loss_mult_minute = 10.0
 

resources/src/ai/trainer.py

@@ -81,16 +81,16 @@ def save_model(self, save_model_file, save_config_file):
         new_model_config = configparser.ConfigParser()
         new_model_config.add_section('Columns')
         columns_section = new_model_config['Columns']
-        columns_section['METRICS'] = ', '.join(self.METRICS)
-        columns_section['TIMESTAMP'] = ', '.join(self.TIMESTAMP)
+        columns_section['METRICS'] = ', '.join(self.metrics)
+        columns_section['TIMESTAMP'] = ', '.join(self.timestamp)
         new_model_config.add_section('General')
         general_section = new_model_config['General']
-        general_section['AVG_LOSS'] = str(self.AVG_LOSS)
-        general_section['STD_LOSS'] = str(self.STD_LOSS)
-        general_section['WINDOW_SIZE'] = str(self.WINDOW_SIZE)
-        general_section['NUM_WINDOWS'] = str(self.NUM_WINDOWS)
-        general_section['LOSS_MULT_1'] = str(self.LOSS_MULT_1)
-        general_section['LOSS_MULT_2'] = str(self.LOSS_MULT_2)
+        general_section['AVG_LOSS'] = str(self.avg_loss)
+        general_section['STD_LOSS'] = str(self.std_loss)
+        general_section['WINDOW_SIZE'] = str(self.window_size)
+        general_section['NUM_WINDOWS'] = str(self.num_window)
+        general_section['LOSS_MULT_METRIC'] = str(self.loss_mult_metric)
+        general_section['LOSS_MULT_MINUTE'] = str(self.loss_mult_minute)
         with open(save_config_file, 'w') as configfile:
             new_model_config.write(configfile)
 
@@ -101,7 +101,7 @@ def data_augmentation(self, data):
 
         Args:
             data (numpy ndarray): original data to be fed to the model.
-        
+
         Returns:
             augmented (numpy ndarray): augmented data.
         """
@@ -113,8 +113,9 @@ def prepare_data_for_training(self, data, augment = False):
 
         Args:
             data (numpy ndarray): data to be used for training.
+
             augment (boolean): set to True to generate more data for training.
-        
+
         Returns:
             prep_data (numpy ndarray): transformed data for its use in the model.
         """
@@ -145,6 +146,6 @@ def train(self, raw_data, epochs=20, batch_size=32, backup_path=None):
         prep_data = self.prepare_data_for_training(data)
         self.model.fit(x=prep_data, y=prep_data, epochs = epochs, batch_size = batch_size, verbose = 0)
         loss = self.model_loss(prep_data, self.model.predict(prep_data), single_value=False).numpy()
-        self.AVG_LOSS = 0.9*self.AVG_LOSS + 0.1*loss.mean()
-        self.STD_LOSS = 0.9*self.AVG_LOSS + 0.1*loss.std()
+        self.avg_loss = 0.9*self.avg_loss + 0.1*loss.mean()
+        self.std_loss = 0.9*self.std_loss + 0.1*loss.std()
         self.save_model(self.model_file ,self.model_config_file)

resources/src/requirements.txt

@@ -35,7 +35,7 @@ pytz==2023.3.post1
 requests==2.31.0
 requests-oauthlib>=1.3.0,<1.4.0
 rsa==4.9
-scikit-learn==1.0.2
+scikit-learn>=1.1.3
 scipy==1.10.0
 six==1.16.0
 tensorboard<2.13,>=2.12