Tushar Tiwari mrtushartiwari

## Base_line.py
median_train_loss = y_train.median()
median_train_loss = [median_train_loss] * len(y_test)
list_shift = [SHIFT] * len(y_test)
print('Mean absolute error on test data for baseline mode ' + str(mean_absolute_error( np.exp( np.array(median_train_loss) - np.array(list_shift)), np.exp(y_test - SHIFT ))))

## featurization_allstate.py
def encode_continous(df,continous_features):
    for col in continous_features:
	df[col +'_log'] = np.log1p(df[col])  # Log transformed
	df[col + '_squareroot'] = np.sqrt(df[col])  # Square root
        df[col + '_square'] = np.square(df[col])  # Square
        df[col + '_log2'] = np.log2(df[col])  # log2

def encode_category(df,categorical_features):
    for col in categorical_features:
        unique_classes = sorted(df[col].unique())

## Boxplot_of_categorical.py
# ref : https://seaborn.pydata.org/tutorial/categorical.html

fig, axs = plt.subplots(1,3,figsize = (22,10) )

sns.boxplot(x="cat80", y="loss",data = train_data, ax= axs[0])
sns.boxplot(x="cat79", y="loss",data = train_data,  ax= axs[1])
sns.boxplot(x="cat87", y="loss",data = train_data,  ax= axs[2])

## counter_barplot.py
#ref : https://stackoverflow.com/a/59840744/5305748  for anotating bargraph
from collections import Counter
k = Counter(counts_categorical)
k = dict(k)
k = pd.DataFrame({'number_of_classes':list(k.keys()),'counts': list(k.values()) })
sns.set(rc={'figure.figsize':(11.0,8.0)})
g = sns.barplot(x= 'number_of_classes', y='counts',data=k)

for p in g.patches:
	g.annotate('{:.0f}'.format(p.get_height()), (p.get_x()+0.3, p.get_height()))

## listofmodels.csv

          
            Model used
             MAE

            
              Linear model Ridge with alpha (0.01)
              1282.15

            
              Linear model Lasso  with alpha (0.01)
               1342.75

            
              Decision Tree Regressor with (max_depth= 12  max_features= n_features  min_samples_leaf= 27)
              1272.19

            
              Ada Boost Regressor with (n_estimators=100  learning_rate=0.0001)
               1342.750

            
              RandomForest Regressor with n_estimators=150  max_features = 100  max_depth = 11
               1342.750

            
              Custom Ensemble model with number of base estimator = 100 
               1209.61

## custom_ensemble.py
def custom_estimator(X_train,y_train, X_test,n_estimators):
	X_train_param = X_train.copy()
	y_train_param = y_train.copy()
    X_test_param = X_test.copy()
    D1_train,D2_train,D1_label,D2_label, =  train_test_split(X_train_param,y_train_param, test_size=0.5 , random_state=42)
    # Combine D1_train and D1_label then sample
    D1_train = D1_train.assign(loss = D1_label )
    base_models = []

    for i in tqdm(range(n_estimators)):

## fetch_data.py
def fetch_data(symbol):
    """ Function to save the Open Low high close volume and adjusted_close data. """
    df = pdr.DataReader(symbol + ".NS", 'yahoo', start='2010-01-01', end='2021-06-30')
    df.to_csv("stock_data/" + symbol) # Saving to a csv file

## selection_stocks.py
f_and_o = pd.read_csv("FandO.csv") # File containing the F&O stocks.

stock_selected = []
for i in f_and_o['Symbol'].values:
    stock_selected.append(i)

# Removing the index future and option contracts
stock_selected.remove('NIFTY')
stock_selected.remove('BANKNIFTY')
stock_selected.remove('FINNIFTY')

## fetching_stock_prices.py
for i in stock_selected:
    fetch_data(symbol = i)
    time.sleep(10)
    print(i)
print(len(os.listdir("stock_data"))) # 156 stocks data downloaded

## istumor.py
def istumor(mask_img):
    """
    Finding wheather the mask has tumor or not
    """
    mask_img = cv2.imread(mask_img)
    k = np.max(mask_img)
    return 1 if k > 0 else 0


images_df["is_tumor"] = images_df['MRI_mask'].apply(istumor)
	median_train_loss = y_train.median()
	median_train_loss = [median_train_loss] * len(y_test)
	list_shift = [SHIFT] * len(y_test)
	print('Mean absolute error on test data for baseline mode ' + str(mean_absolute_error( np.exp( np.array(median_train_loss) - np.array(list_shift)), np.exp(y_test - SHIFT ))))
	def encode_continous(df,continous_features):
	for col in continous_features:
	df[col +'_log'] = np.log1p(df[col]) # Log transformed
	df[col + '_squareroot'] = np.sqrt(df[col]) # Square root
	df[col + '_square'] = np.square(df[col]) # Square
	df[col + '_log2'] = np.log2(df[col]) # log2

	def encode_category(df,categorical_features):
	for col in categorical_features:
	unique_classes = sorted(df[col].unique())
	# ref : https://seaborn.pydata.org/tutorial/categorical.html

	fig, axs = plt.subplots(1,3,figsize = (22,10) )

	sns.boxplot(x="cat80", y="loss",data = train_data, ax= axs[0])
	sns.boxplot(x="cat79", y="loss",data = train_data, ax= axs[1])
	sns.boxplot(x="cat87", y="loss",data = train_data, ax= axs[2])
	#ref : https://stackoverflow.com/a/59840744/5305748 for anotating bargraph
	from collections import Counter
	k = Counter(counts_categorical)
	k = dict(k)
	k = pd.DataFrame({'number_of_classes':list(k.keys()),'counts': list(k.values()) })
	sns.set(rc={'figure.figsize':(11.0,8.0)})
	g = sns.barplot(x= 'number_of_classes', y='counts',data=k)

	for p in g.patches:
	g.annotate('{:.0f}'.format(p.get_height()), (p.get_x()+0.3, p.get_height()))
	Model used	MAE
	Linear model Ridge with alpha (0.01)	1282.15
	Linear model Lasso with alpha (0.01)	1342.75
	Decision Tree Regressor with (max_depth= 12 max_features= n_features min_samples_leaf= 27)	1272.19
	Ada Boost Regressor with (n_estimators=100 learning_rate=0.0001)	1342.750
	RandomForest Regressor with n_estimators=150 max_features = 100 max_depth = 11	1342.750
	Custom Ensemble model with number of base estimator = 100	1209.61
	def custom_estimator(X_train,y_train, X_test,n_estimators):
	X_train_param = X_train.copy()
	y_train_param = y_train.copy()
	X_test_param = X_test.copy()
	D1_train,D2_train,D1_label,D2_label, = train_test_split(X_train_param,y_train_param, test_size=0.5 , random_state=42)
	# Combine D1_train and D1_label then sample
	D1_train = D1_train.assign(loss = D1_label )
	base_models = []

	for i in tqdm(range(n_estimators)):
	def fetch_data(symbol):
	""" Function to save the Open Low high close volume and adjusted_close data. """
	df = pdr.DataReader(symbol + ".NS", 'yahoo', start='2010-01-01', end='2021-06-30')
	df.to_csv("stock_data/" + symbol) # Saving to a csv file
	f_and_o = pd.read_csv("FandO.csv") # File containing the F&O stocks.

	stock_selected = []
	for i in f_and_o['Symbol'].values:
	stock_selected.append(i)

	# Removing the index future and option contracts
	stock_selected.remove('NIFTY')
	stock_selected.remove('BANKNIFTY')
	stock_selected.remove('FINNIFTY')
	for i in stock_selected:
	fetch_data(symbol = i)
	time.sleep(10)
	print(i)
	print(len(os.listdir("stock_data"))) # 156 stocks data downloaded
	def istumor(mask_img):
	"""
	Finding wheather the mask has tumor or not
	"""
	mask_img = cv2.imread(mask_img)
	k = np.max(mask_img)
	return 1 if k > 0 else 0


	images_df["is_tumor"] = images_df['MRI_mask'].apply(istumor)