iRahulPandey

# Database and Table used for feature tables
database_name = "My_DB_Name"
table_name = "titanic_dataset_feature"

# initiate
from databricks import feature_store
fs = feature_store.FeatureStoreClient()

import random

iRahulPandey

from pyspark.ml.feature import StringIndexer
from pyspark.ml.feature import OneHotEncoder
from pyspark.ml.functions import vector_to_array

def create_features(dataframe):
    # take a log of fare
    dataframe = dataframe.withColumn("logFare", f.log10(dataframe.Fare))
    # family head count
    dataframe = dataframe.withColumn("Family", dataframe.SibSp+dataframe.Parch+1)  
    #Average ticket price

iRahulPandey

def get_latest_model_version(model_name):
    latest_version = 1
    mlflow_client = MlflowClient()
    for mv in mlflow_client.search_model_versions(f"name='{model_name}'"):
      version_int = int(mv.version)
      if version_int > latest_version:
        latest_version = version_int
    return latest_version
 
# just passing the test raw data which doesn't have feture columns

iRahulPandey

import csv
# Create a file and open a connection
OUT_FILE = 'bayes_test.csv'
of_connection = open(OUT_FILE, 'w')
writer = csv.writer(of_connection)

ITERATION = 0

# Write column names
headers = ['loss', 'std_loss', 'hyperparameters', 'iteration', 'runtime', 'score']

iRahulPandey

# create a a new search space comparable to grid search and random search and run it for 60 iterations
search_space = {
    'boosting_type': hp.choice('boosting_type', 
                                            [{'boosting_type': 'gbdt', 'subsample': hp.uniform('gdbt_subsample', 0.5, 0.51)}]),
    'subsample_for_bin': hp.uniform('subsample_for_bin', 20000, 20001),
    'min_child_samples': hp.uniform('min_child_samples', 20, 21),
    'num_leaves': hp.uniform('num_leaves', 2, 20),
    'n_estimators': hp.quniform('n_estimators', 1000, 10000, 1000)
}

iRahulPandey

from hyperopt import STATUS_OK
from timeit import default_timer as timer


def objective(hyperparameters):
    
    # Keep track of evals
    global ITERATION
    
    ITERATION += 1

iRahulPandey

# library
from sklearn.model_selection import RandomizedSearchCV

# use repeatedfold from sklearn library
model = lgb.LGBMRegressor()
cv = RepeatedKFold(n_splits=3, n_repeats=3, random_state=0)

# train model
# note: RandomizedSearchCV is maximizing the scoring, hence it is used neg_rmse
# Here i Will use 60 itration for Random search 

iRahulPandey

import random

# create function for random search
def random_search(param_grid, max_evals = MAX_EVALS):

    # Dataframe for results
    results = pd.DataFrame(columns = ['score', 'params', 'iteration'],
                                  index = list(range(MAX_EVALS)))
    
    # Keep searching until reach max evaluations

iRahulPandey

# use gridsearchcv from sklearn library
model = lgb.LGBMRegressor()
cv = RepeatedKFold(n_splits=3, n_repeats=3, random_state=0)

# train model
# note: GridSearchCV is maximizing the scoring, hence it is used neg_rmse
cv_results = GridSearchCV(model, search_space, scoring='neg_root_mean_squared_error', cv=cv).fit(X_train,y_train)

iRahulPandey

# import
from sklearn.model_selection import GridSearchCV, RepeatedKFold

# create a small version of parameter grid defined in Domain and name it as search space - Taking best parameter with little variation in number of leaves and number of estimators
search_space = {
    'boosting_type': ['gbdt'],
    'num_leaves': [2, 4, 6, 8, 10, 12, 14, 16, 18, 20],
    'learning_rate': [0.005],
    'subsample_for_bin': [20000],
    'min_child_samples': [15],