# 'object' columns to dummies
# In[56]:
# 3.0 Read previous application data first
ins = pd.read_csv(
'installments_payments.csv.zip',
nrows = num_rows
)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
ins = reducing.Reducer().reduce(ins)
# In[57]:
# 3.1
ins.shape # (13605401, 8)
ins.head()
# In[58]:
# 3.2 No object type column
ins.dtypes.value_counts()
# 'object' columns to dummies
# In[89]:
# 3.0 Read previous application data first
df = pd.read_csv(
'application_train.csv.zip',
nrows = num_rows
)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
df = reducing.Reducer().reduce(df)
# In[90]:
# 3.0 Read previous application data first
test_df = pd.read_csv(
'application_test.csv.zip',
nrows = num_rows
)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
test_df = reducing.Reducer().reduce(test_df)
# In[44]:
# 2.2 Some constants
num_rows = None # Implies read all rows
nan_as_category = True # While transforming
# 'object' columns to dummies
# In[45]:
# 3.0 Read previous application data first
prev = pd.read_csv('previous_application.csv.zip', nrows=num_rows)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
prev = reducing.Reducer().reduce(prev)
# In[46]:
# 3.0.2
prev.shape # (rows=16,70,214, cols = 37)
prev.head(5)
prev.columns
# In[47]:
# 3.1 Let us examine how many unique IDs exist
prev['SK_ID_PREV'].nunique() # 1670214 Unique number
prev['SK_ID_CURR'].nunique() # 338857 So a number of repeat exist
# We have to aggregate over it
# In[92]:
# 2.1 Some constants
num_rows = None # Implies read all rows
nan_as_category = True # While transforming
# 'object' columns to dummies
# In[93]:
# 3.0 Read previous application data first
df = pd.read_csv('processed_df.csv.zip', nrows=num_rows)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
df = reducing.Reducer().reduce(df)
# In[94]:
# 3.1
df.shape # (356251, 262)
df.head(2)
# In[95]:
# 3.2
df.columns
df.drop(columns=['Unnamed: 0', 'index'], inplace=True)
df.columns
# In[96]:
# 'object' columns to dummies
# In[66]:
# 3.2 Read bureau data first
bureau = pd.read_csv(
'bureau.csv.zip',
nrows = None # Read all rows
)
# 3.2.1 Reduce memory usage by appropriately
# changing data-types per feature:
bureau = reducing.Reducer().reduce(bureau)
# In[67]:
# 3.2.2 Explore data now
bureau.head(5)
bureau.shape # (rows:17,16,428, cols: 17)
bureau.dtypes
# In[68]:
# 3.2.3 In all, how many are categoricals?
num_rows = None # Implies read all rows
nan_as_category = True # While transforming
# 'object' columns to dummies
# ## About the data
# <blockquote>POS_CASH_BALANCE: Monthly data about previous point of sale or cash loans clients have had with <u>Home Credit</u>. Each row is <i>one month</i> of a previous point of sale or cash loan, and a single previous loan can have many rows. This dataset contrasts with <i>bureau_balance</i> dataset where monthly installments were of loans with <u>bureau</u>.</blockquote>
# In[29]:
# 3.2 Read previous application data first
pos = pd.read_csv('POS_CASH_balance.csv.zip', nrows=num_rows)
# 3.0.1 Reduce memory usage by appropriately
# changing data-types per feature:
pos = reducing.Reducer().reduce(pos)
# In[30]:
# 3.3
pos.shape # (rows: 1,00,01358, cols: 8)
pos.head()
# ## Feature explanations
# SK_ID_PREV : ID of previous credit in Home Credit related to loan in our sample. (One loan in our sample can have 0,1,2 or more previous loans in Home Credit)
# SK_ID_CURR: ID of loan in our sample
# MONTHS_BALANCE: Month of balance relative to application date (-1 means the information to the freshest monthly snapshot, 0 means the information at application - often it will be the same as -1 as many banks are not updating the information to Credit Bureau regularly )
# CNT_INSTALMENT: Term of previous credit (can change over time)
# CNT_INSTALMENT_FUTURE: Installments left to pay on the previous credit
# NAME_CONTRACT_STATUS: Contract status during the month
# SK_DPD: DPD (days past due) during the month of previous credit