pwc-india's picture
Update pages/2DATA PROFILER.py
2268e58 verified
raw
history blame
51.2 kB
import streamlit as st
import numpy as np
import pandas as pd
import re
from streamlit_extras.dataframe_explorer import dataframe_explorer
import warnings
from sdv.metadata import SingleTableMetadata
from streamlit_extras.stateful_button import button
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.pipeline import Pipeline
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dense, LSTM, Bidirectional, Conv1D, MaxPooling1D, Flatten, Concatenate, Reshape, RepeatVector
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.losses import MeanSquaredError
from streamlit_extras.stylable_container import stylable_container
from ydata_profiling import ProfileReport
from streamlit_pandas_profiling import st_profile_report
import base64
from sdv.datasets.local import load_csvs
import pyodbc
import pymssql
from streamlit_app import sidebar
warnings.filterwarnings('ignore')
st.set_page_config(
page_title='Profilify: Your AI Assisted Data Profiling App',
layout='wide',
initial_sidebar_state='collapsed'
)
st.markdown("""
<style>
/* Remove blank space at top and bottom */
.block-container {
padding-top: 2.8rem;
/*padding-bottom: 1rem;*/
}
/* Remove blank space at the center canvas */
.st-emotion-cache-z5fcl4 {
position: relative;
top: -62px;
}
/* Make the toolbar transparent and the content below it clickable */
.st-emotion-cache-18ni7ap {
pointer-events: none;
background: rgb(255 255 255 / 0%)
}
.st-emotion-cache-zq5wmm {
pointer-events: auto;
background: rgb(255 255 255);
border-radius: 5px;
}
</style>
""", unsafe_allow_html=True)
def load_dataframe_to_sqlserver(df, table_name, connection_string):
# Establish a connection to the database
conn = pyodbc.connect(connection_string)
cursor = conn.cursor()
# Drop table if it exists
drop_table_sql = f"IF OBJECT_ID('{table_name}', 'U') IS NOT NULL DROP TABLE {table_name}"
try:
cursor.execute(drop_table_sql)
conn.commit()
except Exception as e:
st.error(f"Error dropping table. Please try with a different name.")
# Create table SQL statement based on DataFrame columns and types
create_table_sql = f"CREATE TABLE {table_name} ("
for column in df.columns:
dtype = str(df[column].dtype)
sql_dtype = 'NVARCHAR(MAX)'
create_table_sql += f"{column} {sql_dtype}, "
create_table_sql = create_table_sql.rstrip(', ') + ')'
try:
# Execute table creation
cursor.execute(create_table_sql)
conn.commit()
except Exception as e:
st.error(f"Error Creating table. Please try with a different name.")
# Insert DataFrame data into the table using bulk insert
insert_sql = f"INSERT INTO {table_name} ({', '.join(df.columns)}) VALUES ({', '.join(['?' for _ in df.columns])})"
try:
# Using `fast_executemany` for bulk inserts
cursor.fast_executemany = True
cursor.executemany(insert_sql, df.values.tolist())
conn.commit()
st.success(f"Data Imported with table name: '{table_name}' successfully.")
except Exception as e:
st.error(f"Error Inserting Data. Please try with a different name.")
cursor.close()
conn.close()
def clear_cache():
keys = list(st.session_state.keys())
for key in keys:
st.session_state.pop(key)
def set_bg_hack(main_bg):
'''
A function to unpack an image from root folder and set as bg.
Returns
-------
The background.
'''
# set bg name
main_bg_ext = "png"
st.markdown(
f"""
<style>
.stApp {{
background: url(data:image/{main_bg_ext};base64,{base64.b64encode(open(main_bg, "rb").read()).decode()});
background-size: cover
}}
</style>
""",
unsafe_allow_html=True
)
#set_bg_hack("bg2.png")
header_style = """
<style>
.header {
color: black; /* Soft dark gray text color for readability */
width: 103%;
font-size: 60px; /* Large font size */
font-weight: bold; /* Bold text */
line-height: 1.2; /* Improved readability */
margin-bottom: 30px; /* Add some space below the header */
padding: 20px; /* Add padding for better spacing */
background-image:
linear-gradient(to right, rgba(255, 140, 0, 0.3) 25%, transparent 75%), /* Darker orange with higher opacity */
linear-gradient(to bottom, rgba(255, 140, 0, 0.3) 15%, transparent 75%),
linear-gradient(to left, rgba(255, 140, 0, 0.3) 25%, transparent 55%),
linear-gradient(to top, rgba(255, 140, 0, 0.3) 25%, transparent 95%);
background-blend-mode: overlay;
background-size: 250px 350px;
border-radius: 10px; /* Add border radius for rounded corners */
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); /* Add shadow for depth */
}
</style>
"""
content_style = """
<style>
.content {
font-size: 40px; /* Larger font size for content */
line-height: 1.6; /* Improved readability */
width: 103%;
padding: 10px; /* Add padding for better spacing */
margin-bottom: 20px;
background-color: sky-blue; /* Background color for the header */
border-radius: 10px; /* Add border radius for rounded corners */
box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); /* Add shadow for depth */
}
</style>
"""
small_style = """
<style>
.small {
color: black;
font-size: 30px; /* Larger font size for content */
line-height: 1.6; /* Improved readability */
width: 100%;
padding: 10px; /* Add padding for better spacing */
margin-bottom: 10px;
background-color: white; /* Background color for the header */
border-radius: 10px; /* Add border radius for rounded corners */
}
</style>
"""
def update_column_dtype(df, column_name, dtype):
error_entries = pd.DataFrame()
flag = None
if dtype == 'System Detected':
pass
elif dtype == 'int64':
try:
df[column_name] = df[column_name].astype('int64')
except ValueError:
error_entries = df[~df[column_name].apply(lambda x: str(x).isdigit())]
st.error('Unable to convert some entries to integer. Please Clean the column.')
elif dtype == 'float64/numeric':
try:
df[column_name] = df[column_name].astype('float64')
except ValueError:
error_entries = df[pd.to_numeric(df[column_name], errors='coerce').isna()]
st.error('Unable to convert some entries to float. Please Clean the column.')
elif dtype == 'id':
try:
df[column_name] = df[column_name].astype('int64')
except ValueError:
error_entries = df[~df[column_name].apply(lambda x: str(x).isdigit())]
st.error('Unable to convert some entries to id. Please Clean the column.')
elif dtype == 'categorical/string':
df[column_name] = df[column_name].astype('category')
elif dtype == 'datetime':
try:
df[column_name] = pd.to_datetime(df[column_name], errors='raise', infer_datetime_format=True)
except ValueError:
error_entries = df[pd.to_datetime(df[column_name], errors='coerce', infer_datetime_format=True).isna()]
custom_format = st.text_input("Please provide the datetime format (e.g., %Y-%m-%d):")
if custom_format:
try:
df[column_name] = pd.to_datetime(df[column_name], errors='raise', format=custom_format)
except ValueError:
error_entries = df[pd.to_datetime(df[column_name], errors='coerce', format=custom_format).isna()]
st.error('Unable to parse datetime with the provided format. Please Clean the column.')
elif dtype == 'email':
df[column_name] = df[column_name].astype('category')
flag= 'email'
elif dtype == 'phone_number':
df[column_name] = df[column_name].astype('category')
flag= 'phone_number'
return df, error_entries, flag
def convert_to_special_representation(value):
value = str(value)
special_chars = set("!@#$%^&*()_+-=[]{}|;:,.<>?`~")
result = ''
for char in value:
if char.isdigit():
result += 'N'
elif char.isalpha():
result += 'A'
elif char in special_chars:
result += char
else:
# Handle other characters as needed
result += char
return result
######
def main():
# st.title('PAGE TITLE') # Change this for each page
sidebar()
########
with st.container(border=True):
st.subheader('SELECT TABLE')
metadata = SingleTableMetadata()
# conn = pymssql.connect("Server=sql-ext-dev-uks-001.database.windows.net;"
# "Database=sqldb-ext-dev-uks-001;"
# "UID=dbadmin;"
# "PWD=mYpa$$w0rD" )
conn = pymssql.connect(server="sql-ext-dev-uks-001.database.windows.net"
database="sqldb-ext-dev-uks-001"
user="dbadmin"
password="mYpa$$w0rD" )
query1_1="select * from INFORMATION_SCHEMA.TABLES where TABLE_SCHEMA='dbo' and TABLE_NAME in ('TCM', 'TCVM','TEM', 'TPM', 'TPP', 'TPT', 'TRM', 'TSCM', 'TSM') ORDER BY TABLE_NAME ASC"
query1_2="select * from INFORMATION_SCHEMA.TABLES where TABLE_SCHEMA='dbo' and TABLE_NAME LIKE 'PROFILED%' ORDER BY TABLE_NAME ASC"
tab_names=list(pd.read_sql_query(query1_1,con=conn)['TABLE_NAME'])
tab_names_edited= list(pd.read_sql_query(query1_2,con=conn)['TABLE_NAME'])
sample_selector=st.selectbox('SELECT SAMPLE SIZE',['100','10K','100K','1M','Full Table'],index=None,placeholder='Select sample size for the table(s)', on_change= clear_cache)
mode_selector=st.selectbox("Select How you want to Proceed", ["Start Profiling with Source Data", "Load Previously Profiled Data For Further Processing"], on_change=clear_cache,placeholder='Show Options')
if mode_selector == "Start Profiling with Source Data":
table_selector=st.selectbox('SELECT TABLE NAME',tab_names,index=None,on_change=clear_cache,placeholder='Select table name')
if mode_selector == "Load Previously Profiled Data For Further Processing":
table_selector=st.selectbox('SELECT TABLE NAME',tab_names_edited,index=None,on_change=clear_cache,placeholder='Select table name')
if table_selector is not None and sample_selector is not None:
if sample_selector=='100':
count="top 100"
elif sample_selector=='10K':
count="top 10000"
elif sample_selector=='100K':
count="top 100000"
elif sample_selector=='1M':
count="top 1000000"
else:
count=""
query2="select "+count+" * from [dbo].["+table_selector+"]"
df = pd.read_sql_query(query2,con=conn)
main_list=df.columns.to_list()
sub_list=['ID','LOADID','FILE_NAME']
if any(main_list[i:i+len(sub_list)] == sub_list for i in range(len(main_list) - len(sub_list) + 1)):
df=df.drop(['ID','LOADID','FILE_NAME'],axis=1)
conn.close()
if 'data' not in st.session_state:
st.session_state.data= df
metadata.detect_from_dataframe(st.session_state.data)
st.sidebar.header("DataFrame Live Preview")
st.sidebar.markdown("*This Window keeps the live status of the dataframe under processing. You can review this dataframe after all the changes.*")
df_preview= st.sidebar.empty()
df_preview.write(st.session_state.data)
st.markdown(content_style, unsafe_allow_html=True)
with st.container(border=True):
cols= df.columns.to_list()
primary_key= metadata.primary_key
sugg_primary_keys = [col for col in cols if df[col].is_unique and df[col].dtype != 'float' and not df[col].isnull().any()]
prob_key= sugg_primary_keys
if primary_key in sugg_primary_keys:
default_index = sugg_primary_keys.index(primary_key)
else:
sugg_primary_keys.append(primary_key)
default_index = sugg_primary_keys.index(primary_key)
no_y_data =[]
email_cols=[]
phone_cols=[]
# cols_select= st.multiselect('Please select column(s) for Profiling and Cleansing', cols, default= cols[:5])
tabs3= st.tabs(cols)
for i, tab in enumerate(tabs3):
with tab:
col= cols[i]
scol1,scol2= st.columns([4,1])
with scol1:
taba, tabb, tabc, tabd, tabe = st.tabs(["📝 DataType Validation", "🧹 Missing Value Handling", "📈 Statistical Profiling", " ✨ Pattern Exploration", "🤖 AI Assisted Data Cleansing"])
with taba:
if st.session_state.data[col].dtype.name == 'category':
st.session_state.data[col] = st.session_state.data[col].astype('str')
dtypes= ['System Detected', 'int64', 'float64/numeric', 'id', 'categorical/string','datetime', 'email', 'phone_number']
no_dtypes= ['int64', 'float64/numeric', 'id', 'categorical/string','datetime', 'email', 'phone_number']
no_dtype = False
if metadata.columns[col]['sdtype'] != "unknown":
datatype= metadata.columns[col]['sdtype']
st.info(f"System Identified DataType: {datatype}")
elif str(df[col].dtype) != 'object' and metadata.columns[col]['sdtype'] == "unknown":
datatype= str(df[col].dtype)
st.info(f"System Identified DataType: {datatype}")
else:
datatype= 'NA'
#st.warning("System Could Not Understand Datatype. Please Specify the Datatype")
no_dtype= True
if datatype in ['int64']:
def_index=1
if datatype in ['float64', 'numerical']:
def_index=2
if datatype in ['id']:
def_index=3
if datatype in ['categorical', 'string']:
def_index=4
if datatype in ['datetime']:
def_index=5
if datatype in ['email']:
def_index=6
if datatype in ['phone_number']:
def_index=7
if col == primary_key:
st.success("This is System Identified Primary Key")
elif col in prob_key:
st.warning("This is System suggested potential Primary Key")
if f'dtype_{col}' not in st.session_state:
st.session_state[f'dtype_{col}'] = 'initiate'
if st.session_state[f'dtype_{col}'] not in ['email', 'phone_number']:
st.session_state.flag = None
if no_dtype == True:
fin_datatype= st.selectbox(f"Please Change/Define the Datatype of column: {col}:",no_dtypes, index=3, key= f'datatype_{col}')
else:
fin_datatype= st.selectbox(f"Please Change/Define the Datatype of column: {col}:",dtypes, index=def_index, key= f'datatype_{col}')
st.session_state[f'dtype_{col}'] = st.session_state[f'datatype_{col}']
st.session_state.data, error_df, st.session_state.flag= update_column_dtype(st.session_state.data,col,fin_datatype)
if error_df.empty:
st.success("No Datatype Validation Errors For Current Datatype")
try:
df_preview.write(st.session_state.data)
except:
st.warning("DataFrame Updated. But Could Not Load Preview")
else:
st.subheader("Prepare the Column for Conversion:")
try:
edited_error_df= st.data_editor(error_df, num_rows="dynamic",column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, key=f'dtype_error_{col}')
except:
edited_error_df= st.data_editor(error_df, num_rows="dynamic",column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, key=f'dtype_error_{col}')
check = st.button("Fix Error", key=f"Fix{col}")
if check:
st.session_state.data= st.session_state.data.drop(error_df.index)
st.session_state.data = pd.concat([st.session_state.data, edited_error_df])
df_preview.write(st.session_state.data)
if fin_datatype in ['id', 'email', 'phone_number']:
no_y_data.append(col)
if fin_datatype in ['email']:
email_cols.append(col)
if fin_datatype in ['phone_number']:
phone_cols.append(col)
no_y_data.extend(['Validity','Validity_phone','Validity_email'])
total_records = len(st.session_state.data)
with tabc:
if col not in no_y_data:
y_data_col= st.session_state.data[[col]]
pr = ProfileReport(y_data_col, dark_mode=True, explorative=False, config_file=r"ydata_config.yml")
pr.config.html.style.primary_colors = ['#e41a1c']
with st.container(border=True):
st_profile_report(pr, navbar=False, key=f'profile{col}')
elif col in email_cols:
unique_emails = st.session_state.data[col].nunique()
duplicate_emails = total_records - unique_emails
# Extract email domains
email_domains = st.session_state.data[col].str.extract(r'@(.+)$')[0]
# Count occurrences of each domain
email_domain_counts = email_domains.value_counts()
# Get the top 5 email domains
top_email_domains = email_domain_counts.head(5)
# Format the top email domains for display
top_email_domains_str = '\n|\n'.join([f"{domain}: {count}" for domain, count in top_email_domains.items()])
if f'invalid_em_{col}' in st.session_state:
invalid_emails= len(st.session_state[f'invalid_em_{col}'])
valid_emails= total_records - invalid_emails
percent_invalid_emails = invalid_emails / total_records * 100
email_message = f"""
## Email Column: {col}\n\n **Valid Emails:** {valid_emails} ({100 - percent_invalid_emails:.2f}%)\n\n---------------------------------------------------------------------------------------\n\n**Invalid Emails:** {invalid_emails} ({percent_invalid_emails:.2f}%)\n\n----------------------------------------------------------------------------------------\n\n**Unique Emails:** {unique_emails}\n\n-------------------------------------------------------------------------------------------------------------------------\n\n**Duplicate Emails:** {duplicate_emails}\n\n----------------------------------------------------------------------------------------------------------------------\n\n**Top 5 Email Domains:** {top_email_domains_str}
"""
else:
invalid_emails= "Please Execute AI Assisted Data Validation on Email Columns for Profiling Report of them."
valid_emails= "Please Execute AI Assisted Data Validation on Email Columns for Profiling Report of them."
percent_invalid_emails = "Please Execute AI Assisted Data Validation on Email Columns for Profiling Report of them."
email_message = f"""
## Email Column: {col}\n\n **Valid Emails:** {valid_emails} \n\n---------------------------------------------------------------------------------------\n\n**Invalid Emails:** {invalid_emails}\n\n----------------------------------------------------------------------------------------\n\n**Unique Emails:** {unique_emails}\n\n-------------------------------------------------------------------------------------------------------------------------\n\n**Duplicate Emails:** {duplicate_emails}\n\n----------------------------------------------------------------------------------------------------------------------\n\n**Top 5 Email Domains:** {top_email_domains_str}
"""
with st.container(border=True):
st.markdown(str(email_message))
ref_em=st.button('Refresh', key=f'email{col}')
if ref_em:
pass
elif col in phone_cols:
unique_phones = st.session_state.data[col].nunique()
duplicate_phones = total_records - unique_phones
phone_country_codes = st.session_state.data[col].str.extract(r'^\+(\d+)')[0].value_counts()
top_phone_country_codes = list(phone_country_codes.head(5).to_string())
to_remove = ['\n', ' ']
top_phone_country_codes = [item for item in top_phone_country_codes if item not in to_remove]
if f'invalid_ph_{col}' in st.session_state:
invalid_phones= len(st.session_state[f'invalid_ph_{col}'])
valid_phones= total_records - invalid_phones
percent_invalid_phones = invalid_phones / total_records * 100
phone_message= f"""
## Phone Number Column: {col}\n\n **Valid Phone Numbers:** {valid_phones} ({100 - percent_invalid_phones:.2f}%)\n\n----------------------------------------------------------------------------------------------------------\n\n**Invalid Phone Numbers:** {invalid_phones} ({percent_invalid_phones:.2f}%)\n\n----------------------------------------------------------------------------------------------------------\n\n**Unique Phone Numbers:** {unique_phones}\n\n----------------------------------------------------------------------------------------------------------\n\n**Duplicate Phone Numbers:** {duplicate_phones}\n\n----------------------------------------------------------------------------------------------------------\n\n**Top 5 Phone Country Codes:** {top_phone_country_codes}
"""
else:
invalid_phones= "Please Execute AI Assisted Data Validation on Phone Number Columns for Profiling Report of them."
valid_phones= "Please Execute AI Assisted Data Validation on Phone Number Columns for Profiling Report of them."
percent_invalid_phones = "Please Execute AI Assisted Data Validation on Phone Number Columns for Profiling Report of them."
phone_message=f"""
## Phone Number Column: {col}\n\n **Valid Phone Numbers:** {valid_phones} \n\n----------------------------------------------------------------------------------------------------------\n\n **Invalid Phone Numbers:** {invalid_phones} \n\n----------------------------------------------------------------------------------------------------------\n\n **Unique Phone Numbers:** {unique_phones}\n\n----------------------------------------------------------------------------------------------------------\n\n **Duplicate Phone Numbers:** {duplicate_phones}\n\n----------------------------------------------------------------------------------------------------------\n\n **Top 5 Phone Country Codes:** {top_phone_country_codes}
"""
with st.container(border=True):
st.markdown(str(phone_message))
ref_ph=st.button('Refresh', key=f'phone{col}')
if ref_ph:
pass
with tabd:
st.session_state.data_encoded = st.session_state.data.copy()
st.session_state.data_encoded[f'Pattern_{col}'] = st.session_state.data_encoded[col].apply(convert_to_special_representation)
patterns= list(st.session_state.data_encoded[f'Pattern_{col}'].unique())
patt_col1, patt_col2 = st.columns([1,4])
with patt_col1:
st.session_state.pattern_list= pd.DataFrame(patterns,columns=['Pattern Name'])
event = st.dataframe(
st.session_state.pattern_list,
key=f"pattern_list_data{col}",
on_select="rerun",
selection_mode=["multi-row"],
hide_index=True,
width= 10000,
height= 450
)
if len(event.selection.rows) > 0:
filter= list(st.session_state.pattern_list.loc[event.selection.rows]['Pattern Name'].values)
else:
filter = None
if filter is not None:
with patt_col2:
with st.container(border= True, height= 450):
st.write("#####")
if not st.session_state.data_encoded[st.session_state.data_encoded[f'Pattern_{col}'].isin(filter)].empty:
st.session_state.data_encoded[col] = st.session_state.data_encoded[col].astype('str')
try:
edited_pattern_df= st.data_editor(st.session_state.data_encoded[st.session_state.data_encoded[f'Pattern_{col}'].isin(filter)], num_rows="dynamic",column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, height=300, key=f'Valid_pattern_{col}')
except:
edited_pattern_df= st.data_editor(st.session_state.data_encoded[st.session_state.data_encoded[f'Pattern_{col}'].isin(filter)], num_rows="dynamic",column_config={
col: st.column_config.Column(
col,
width="medium",
)
}, height=300, key=f'Valid_pattern_{col}')
valid_pattern = st.button("Confirm", key=f"Fix_valid_pattern_{col}")
if valid_pattern:
st.session_state.data= st.session_state.data.drop(st.session_state.data_encoded[st.session_state.data_encoded[f'Pattern_{col}'].isin(filter)].index)
st.session_state.data = pd.concat([st.session_state.data, edited_pattern_df])
st.session_state.data=st.session_state.data.drop([f'Pattern_{col}'], axis=1)
st.session_state.data= st.session_state.data.sort_index()
df_preview.write(st.session_state.data)
else:
with patt_col2:
with stylable_container(
key=f"container_select_pattern_none{col}",
css_styles="""
{
border: 1px solid white;
border-radius: 0.5rem;
padding: calc(1em - 1px);
width: 100%;
color: orange;
size: 100px;
}
"""
):
st.write('##\n\n##\n\n')
st.markdown("""
<style>
.big-font {
font-size:15px;
width: 100%;
text-align: center;
}
</style>
""", unsafe_allow_html=True)
st.markdown(f'<p class="big-font">🛈 There are total {len(st.session_state.pattern_list)} Number of Patterns Available. Please Select Pattern(s) for Matching Records</p>', unsafe_allow_html=True)
st.write('##\n\n##\n\n')
with tabb:
try:
edited_df= st.data_editor(st.session_state.data[(st.session_state.data[col].isna()) | (st.session_state.data[col] == '') | (st.session_state.data[col] == None)], num_rows="dynamic", column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, key=f'miss_{col}')
except:
edited_df= st.data_editor(st.session_state.data[(st.session_state.data[col].isna()) | (st.session_state.data[col] == '') | (st.session_state.data[col] == None)], num_rows="dynamic", column_config={
col: st.column_config.Column(
col,
width="medium",
)
}, key=f'miss_{col}')
incol1,incol2, extra= st.columns([1.1,1.5,8])
with incol1:
#st.write(st.session_state[f'dtype_{col}'])
if st.session_state[f'dtype_{col}'] not in ['int64', 'float64/numeric']:
def_fill = st.text_input("Default Autofill Value",key=f"def_fill_{col}")
autofill= st.button("Autofill", key=f"autofill_{col}")
if autofill:
if st.session_state[f'dtype_{col}'] not in ['int','float']:
st.session_state.data[col] = st.session_state.data[col].astype('str').replace('', pd.NA).replace({None: pd.NA}).fillna(def_fill)
else:
st.session_state.data[col] = st.session_state.data[col].replace({None: pd.NA}).fillna(method='ffill')
st.success("Column Autofilled. Please Review the Sidebar for updated status of the Dataframe.")
df_preview.write(st.session_state.data)
with incol2:
confirm= st.button("Confirm", key=f"Confirm_{col}")
if confirm:
st.session_state.data[col] = st.session_state.data[col].replace('', np.nan).replace({None: np.nan})
st.session_state.data = st.session_state.data.dropna(subset=[col])
st.session_state.data.update(edited_df)
st.session_state.data = pd.concat([st.session_state.data, edited_df[~edited_df.index.isin(st.session_state.data.index)]])
st.session_state.data= st.session_state.data.sort_index()
st.success("State Saved. Please Review the Sidebar for updated status of the Dataframe.")
df_preview.write(st.session_state.data)
with tabe:
if "overall_invalid_df" not in st.session_state:
st.session_state.overall_invalid_df = pd.DataFrame()
if (st.session_state[f'dtype_{col}'] not in ['email', 'phone_number'] and st.session_state.flag not in ['email', 'phone_number']):
st.dataframe(st.session_state.data)
AI_check= st.button("Check For Anomalies", key= f'AI_CHECK_{col}')
if AI_check:
with st.spinner("Running Anomaly Detection AI"):
#my_bar = st.progress(0, text="Progress")
if st.session_state[f'dtype_{col}'] in ['categorical/string']:
if 'missing@123' not in st.session_state.data[col].cat.categories:
st.session_state.data[col] = st.session_state.data[col].cat.add_categories(['missing@123'])
st.session_state.data[col] = st.session_state.data[col].fillna('missing@123').astype(str)
st.session_state.data_encoded = st.session_state.data[col].apply(convert_to_special_representation)
mixed_transformer = Pipeline(steps=[
('vectorizer', CountVectorizer(analyzer='char', lowercase=False))
])
df_transformed = mixed_transformer.fit_transform(st.session_state.data_encoded)
input_dim = df_transformed.shape[1]
encoding_dim = (input_dim // 2) + 1
input_layer = Input(shape=(None, input_dim))
conv1d_layer = Conv1D(64, 3, activation='relu', padding='same')(input_layer)
maxpooling_layer = MaxPooling1D(pool_size=2, padding='same')(conv1d_layer)
encoder_lstm = Bidirectional(LSTM(encoding_dim, activation='relu', return_sequences=False))(maxpooling_layer)
repeat_vector = RepeatVector(input_dim)(encoder_lstm)
decoder_lstm = Bidirectional(LSTM(encoding_dim, activation='relu', return_sequences=True))(repeat_vector)
conv1d_layer_decoder = Conv1D(64, 3, activation='relu', padding='same')(decoder_lstm)
upsampling_layer = Conv1D(input_dim, 2, activation='relu', padding='same')(conv1d_layer_decoder)
autoencoder = Model(inputs=input_layer, outputs=upsampling_layer)
autoencoder.compile(optimizer=Adam(), loss=MeanSquaredError())
#my_bar.progress(40, text='Progress')
autoencoder.fit(np.expand_dims(df_transformed.toarray(), axis=1), np.expand_dims(df_transformed.toarray(), axis=1),
epochs=100, batch_size=2, shuffle=True, validation_split=0.2, verbose=1)
reconstructions = autoencoder.predict(np.expand_dims(df_transformed.toarray(), axis=1))
reconstruction_error = np.mean(np.abs(reconstructions - np.expand_dims(df_transformed.toarray(), axis=1)), axis=(1, 2))
threshold = np.percentile(reconstruction_error, 95) # Adjust the percentile based on desired sensitivity
#my_bar.progress(90, text='Progress')
st.session_state.data['Validity'] = ['Invalid' if error > threshold else 'Valid' for error in reconstruction_error]
st.session_state.data[col] = st.session_state.data[col].replace('missing@123', '')
st.session_state[f"invalid_ai_data_{col}"]= st.session_state.data[st.session_state.data['Validity']== 'Invalid']
#my_bar.progress(100, text='Progress')
if f"invalid_ai_data_{col}" in st.session_state:
st.session_state[f"invalid_ai_data_{col}"]["Invalid Field"] = col
if 'Validity' in st.session_state[f"invalid_ai_data_{col}"].columns:
st.session_state.overall_invalid_df = pd.concat([st.session_state.overall_invalid_df, st.session_state[f"invalid_ai_data_{col}"].drop(['Validity'], axis=1)], ignore_index=True)
else:
st.session_state.overall_invalid_df = pd.concat([st.session_state.overall_invalid_df, st.session_state[f"invalid_ai_data_{col}"]], ignore_index=True)
try:
edited_valid_df= st.data_editor(st.session_state[f"invalid_ai_data_{col}"], num_rows="dynamic",column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, key=f'Valid_{col}')
except:
edited_valid_df= st.data_editor(st.session_state[f"invalid_ai_data_{col}"], num_rows="dynamic",column_config={
col: st.column_config.Column(
col,
width="medium",
)
}, key=f'Valid_{col}')
valid = st.button("Confirm", key=f"Fix_valid_{col}")
#my_bar.empty()
if valid:
st.session_state.data= st.session_state.data.drop(st.session_state.data[st.session_state.data['Validity'] == 'Invalid'].index)
st.session_state.data = pd.concat([st.session_state.data, edited_valid_df])
st.session_state.data= st.session_state.data.sort_index()
df_preview.write(st.session_state.data)
elif (st.session_state[f'dtype_{col}'] in ['phone_number'] or st.session_state.flag in ['phone_number'] ):
#st.dataframe(st.session_state.data)
phone_regex = r'^\+?[0-9\s\-\(\)]+$'
# st.write(phone_regex)
st.session_state.data['Validity_phone'] = st.session_state.data[col].apply(lambda xy: 'phone_is_valid' if re.match(phone_regex,str(xy)) else 'phone_is_invalid')
st.session_state[f'invalid_phone_{col}']= st.session_state.data[st.session_state.data['Validity_phone'] == 'phone_is_invalid'].drop(['Validity_phone'], axis=1)
if f'invalid_phone_{col}_check' not in st.session_state:
st.session_state[f'invalid_phone_{col}']["Invalid Field"] = col
st.session_state.overall_invalid_df = pd.concat([st.session_state.overall_invalid_df, st.session_state[f'invalid_phone_{col}']], ignore_index=True, axis=0)
st.session_state[f'invalid_phone_{col}_check'] = 'yes'
try:
edited_valid_df= st.data_editor(st.session_state.data[st.session_state.data['Validity_phone'] == 'phone_is_invalid'], column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, num_rows="dynamic", key=f'Valid_phone_{col}')
except:
edited_valid_df= st.data_editor(st.session_state.data[st.session_state.data['Validity_phone'] == 'phone_is_invalid'], column_config={
col: st.column_config.Column(
col,
width="medium",
)
}, num_rows="dynamic", key=f'Valid_phone_{col}')
valid_phone = st.button("Confirm", key=f"Fix_valid_phone_{col}")
if valid_phone:
st.session_state.data= st.session_state.data.drop(st.session_state.data[st.session_state.data['Validity_phone'] == 'phone_is_invalid'].index)
st.session_state.data = pd.concat([st.session_state.data, edited_valid_df])
st.session_state[f'invalid_ph_{col}']= st.session_state.data[st.session_state.data['Validity_phone'] == 'phone_is_invalid'].drop(['Validity_phone'], axis=1)
st.session_state.data = st.session_state.data.drop(['Validity_phone'], axis=1)
df_preview.write(st.session_state.data)
elif (st.session_state[f'dtype_{col}'] in ['email'] or st.session_state.flag in ['email']):
email_regex = r'^[a-zA-Z0-9_.+-]+@[a-zA-Z0-9-]+\.[a-zA-Z0-9-.]+$'
st.session_state.data['Validity_email'] = st.session_state.data[col].apply(lambda x: 'email_is_valid' if re.match(email_regex, x) else 'email_is_invalid')
if st.session_state.data[col].dtype.name == 'category':
st.session_state.data[col] = st.session_state.data[col].astype('str')
st.session_state[f'invalid_email_{col}']= st.session_state.data[st.session_state.data['Validity_email'] == 'email_is_invalid'].drop(['Validity_email'], axis=1)
if f'invalid_email_{col}_check' not in st.session_state:
st.session_state[f'invalid_email_{col}']["Invalid Field"] = col
st.session_state.overall_invalid_df = pd.concat([st.session_state.overall_invalid_df, st.session_state[f'invalid_email_{col}']], ignore_index=True, axis=0)
st.session_state[f'invalid_email_{col}_check'] = 'yes'
try:
edited_valid_df= st.data_editor(st.session_state.data[st.session_state.data['Validity_email'] == 'email_is_invalid'], num_rows="dynamic", column_config={
col: st.column_config.TextColumn(
col,
width="medium",
)
}, key=f'Valid_email_{col}')
except:
edited_valid_df= st.data_editor(st.session_state.data[st.session_state.data['Validity_email'] == 'email_is_invalid'], num_rows="dynamic", column_config={
col: st.column_config.Column(
col,
width="medium",
)
}, key=f'Valid_email_{col}')
valid_email = st.button("Confirm", key=f"Fix_valid_email_{col}")
if valid_email:
st.session_state.data= st.session_state.data.drop(st.session_state.data[st.session_state.data['Validity_email'] == 'email_is_invalid'].index)
st.session_state.data = pd.concat([st.session_state.data, edited_valid_df])
st.session_state[f'invalid_em_{col}']= st.session_state.data[st.session_state.data['Validity_email'] == 'email_is_invalid'].drop(['Validity_email'], axis=1)
st.session_state.data = st.session_state.data.drop(['Validity_email'], axis=1)
df_preview.write(st.session_state.data)
with scol2:
st.markdown("**Column Being Processed**")
col_view= st.empty()
try:
col_view.write(st.session_state.data[col])
except:
st.warning("DataFrame Updated. But Could Not Load Preview")
pkcol1, pkcol2=st.columns(2)
with pkcol1:
if primary_key != None:
st.info(f"Primary Key Identified by AI: {primary_key}")
else:
st.warning("Could Not Finalize the Primary Key Automatically. Please go through the suggestions and Finalize one.")
with pkcol2:
st.selectbox("Please Finalize the Primary Key:", sugg_primary_keys, index= default_index)
with st.expander("Save and Download Data"):
name_data= st.text_input("Please Specify Name of the saved/downloaded data")
csv = st.session_state.data.to_csv(index=False).encode('utf-8')
for col in ['Validity', 'Validity_email', 'Validity_phone']:
if col in st.session_state.overall_invalid_df:
st.session_state.overall_invalid_df = st.session_state.overall_invalid_df.drop([col], axis=1)
csv2 = st.session_state.overall_invalid_df.to_csv(index=False).encode('utf-8')
#st.write(st.session_state.overall_invalid_df)
# Create a download button
dldcol1, dldcol2= st.columns([1,4])
with dldcol1:
st.download_button(
label="Download Cleaned Data as CSV",
data=csv,
file_name=f'{name_data}.csv',
mime='text/csv',
)
with dldcol2:
st.download_button(
label="Download Anomalous Data as CSV",
data=csv2,
file_name=f'Anomaly_{name_data}.csv',
mime='text/csv',
)
save = st.button("Save Data For Further Processing")
if save:
connection_string = ( 'SERVER=sql-ext-dev-uks-001.database.windows.net;'
'DATABASE=sqldb-ext-dev-uks-001;'
'UID=dbadmin;'
'PWD=mYpa$$w0rD'
)
st.session_state.data = st.session_state.data.astype(str)
load_dataframe_to_sqlserver(st.session_state.data, f'[dbo].[PROFILED_{name_data}]', connection_string)
######
if __name__ == '__main__':
main()