Underwriting Model

Installation

Disturbution Tested on

• Ubuntu server 16.04

Requirments

• python3

Install pip3

sudo apt install python3-pip

Clone the repository

git clone https://github.com/mehtamanan0/underwritingModel.git; cd underwritingModel

Install python dependencies

pip3 install -r requirements.txt
python3 -m spacy download en_core_web_md

Running the server

Start the server

./start_server.sh

Once the server is running all the log will be saved to log.txt

Stop the server

./stop_server.sh

Testing

NOTE: change the host address to the address on which the server is running

Using a browser

URL :

http://127.0.0.1:5000/predict?amount_pence=113542&balance_mortgage=0&direct_debit_day=2&estatus=8&gross_income=0&home_value=0&hstatus=1&monthly_mortgage=0&monthly_payment=120.58&net_income=0&number_of_dependants=0&reason=10&reasonother=TravelTicket&rent=0&rsaddress_town=London&rsemployment_employers_name=0&rsemployment_job_title=0&rsemployment_pre_employer_name=0&rsemployment_pre_job_title=0&rsemployment_rsaddress_town=0&work_number=7886870267

Copy the link and paste it into a browser

Using curl

curl "http://127.0.0.1:5000/predict?amount_pence=113542&balance_mortgage=0&direct_debit_day=2&estatus=8&gross_income=0&home_value=0&hstatus=1&monthly_mortgage=0&monthly_payment=120.58&net_income=0&number_of_dependants=0&reason=10&reasonother=TravelTicket&rent=0&rsaddress_town=London&rsemployment_employers_name=0&rsemployment_job_title=0&rsemployment_pre_employer_name=0&rsemployment_pre_job_title=0&rsemployment_rsaddress_town=0&work_number=7886870267"

The output should look something like this :

{
  "Percentage of approval": 0.6109171426229283,
  "percentage of rejection": 0.3890828573770717
}

File Functions and Explanations

predict_server.py

Function

Creates a REST api server for GET requests. It can be accessed by giving it the Api parameters given below. Once it gets the parameters, it will give these parameters to the model which will then predict probabilities and return in a json format.

Explanations

this file will load the model from the models/ directory as well as the vector model here :

nlp = spacy.load('en_core_web_md')
model = joblib.load("models/model")

Then the the function predict will fetch GET request values like so:

amount_pence = float(request.args.get('amount_pence'))

After getting all parameters, all the parameters will be sent to the model for probability prediction:

pred = model.predict_proba([model_values])[0]

After that the request will return the output in a json format

train_model.py

Function

This file reads the preprocessed data and generates a model that gets saved in the models directory

Explanations

The file will first read the preprocessed csv file:

train_df = pd.read_csv("data/train_data_preprocessed.csv")

Then the training data will be split like:

X = train_df.values[:, 0:21].astype(float)

Y = train_df.values[:, 21].astype(int)

Here X contains all the data values for training And Y contains the corresponding labels for the data where:

• 1 - Approved
• 0 - Declined

Then the model will be trained like so:

model.fit(X, Y)

After training the model the model will be saved in the models directory

Api arguments

The api is a GET request with the following parameters:

• amount_pence (int)
• balance_mortgage (int)
• direct_debit_day (int)
• estatus (int)
• gross_income (int)
• home_value (int)
• hstatus (int)
• monthly_mortgage (int)
• monthly_payment (int)
• net_income (int)
• number_of_dependants (int)
• reason (int)
• rent (int)
• work_number (int)
• reasonother (string)
• rsaddress_town (string)
• rsemployment_employers_name (string)
• rsemployment_job_title (string)
• rsemployment_rsaddress_town (string)
• rsemployment_pre_employer_name (string)
• rsemployment_pre_job_title (string)

NOTE

if you get the following error:

Traceback (most recent call last):
  File "predict_server.py", line 11, in <module>
    model = joblib.load("models/model")
  File "/home/manan/.local/lib/python3.5/site-packages/sklearn/externals/joblib/numpy_pickle.py", line 578, in load
    obj = _unpickle(fobj, filename, mmap_mode)
  File "/home/manan/.local/lib/python3.5/site-packages/sklearn/externals/joblib/numpy_pickle.py", line 508, in _unpickle
    obj = unpickler.load()
  File "/usr/lib/python3.5/pickle.py", line 1039, in load
    dispatch[key[0]](self)
  File "/home/manan/.local/lib/python3.5/site-packages/sklearn/externals/joblib/numpy_pickle.py", line 341, in load_build
    self.stack.append(array_wrapper.read(self))
  File "/home/manan/.local/lib/python3.5/site-packages/sklearn/externals/joblib/numpy_pickle.py", line 184, in read
    array = self.read_array(unpickler)
  File "/home/manan/.local/lib/python3.5/site-packages/sklearn/externals/joblib/numpy_pickle.py", line 108, in read_array
    array = pickle.load(unpickler.file_handle)
  File "sklearn/tree/_tree.pyx", line 601, in sklearn.tree._tree.Tree.__cinit__
ValueError: Buffer dtype mismatch, expected 'SIZE_t' but got 'long long'

Run

python3 train_model.py

This is caused due to the model mismatch on different architectures. The above command will recreate the model on current architecture.