Social Indicators Prediction

[1]:

import pandas as pd

data = pd.read_pickle('./../data/social_regression_data.pickle')
data.head()

[1]:
population site_area is_living footprint_area build_floor_area living_area non_living_area buildings_count agriculture industrial ... guest_house_count school_count pitch_count hospital_count polyclinic_count emergency_medical_service_stations_count restaurant_count pharmacy_count longitude latitude
3 2154 958000.0 107.0 134350.198599 1.486228e+05 18790.798548 129551.116050 973 9.895049e+06 4.271492e+06 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -771885.613281 6.809647e+06
4 1698 732000.0 15.0 167346.131884 1.839993e+05 19628.059600 164371.238284 1016 3.137606e+06 2.275153e+05 ... 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 -780323.931010 6.841250e+06
5 20169 45000.0 1484.0 897718.375067 1.446731e+06 530190.233798 899757.274216 3448 6.347738e+04 1.899003e+06 ... 0.0 4.0 9.0 1.0 1.0 0.0 5.0 7.0 -775855.937394 6.821746e+06
6 3393 699000.0 70.0 207072.938797 2.546795e+05 56015.386519 195416.138670 502 1.859899e+06 1.173573e+07 ... 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 -798102.040434 6.809219e+06
7 15960 261000.0 225.0 479686.323942 8.471635e+05 353020.537179 484596.269902 2155 1.301242e+06 2.936352e+05 ... 0.0 3.0 5.0 1.0 1.0 0.0 1.0 3.0 -787756.552231 6.821318e+06

5 rows × 45 columns

Predicting using quantile regression

[3]:

from blocksnet.machine_learning.regression import SocialRegressor

# sr = SocialRegressor()

# if you want to train one yourself
sr = SocialRegressor(n_estimators=100, max_depth=4, min_samples_split=5)

2025-05-27 16:04:58.876 | WARNING  | blocksnet.machine_learning.regression.social.common.model_wrapper:__init__:32 - Keyword arguments provided. The model_path is ignored

Getting data divided into a training and a test sample

[4]:

x_train, x_test, y_train, y_test = sr.get_train_data(data)

Train model

[5]:

sr.train(x_train, y_train, confidence_level=95.0)

2025-05-27 16:05:16.289 | INFO     | blocksnet.machine_learning.regression.social.common.model_wrapper:_train_model:80 - Fitting median model
2025-05-27 16:05:47.337 | INFO     | blocksnet.machine_learning.regression.social.common.model_wrapper:_train_model:89 - Fitting lower model
2025-05-27 16:05:49.852 | INFO     | blocksnet.machine_learning.regression.social.common.model_wrapper:_train_model:100 - Fitting upper model

Save fitted model

[6]:

sr.save_model('./model.pickle')

Model Validation

[7]:

y_pred, pi_lower, pi_upper = sr.evaluate(x_test)

Showing predictions

[8]:

y_pred.head()

[8]:
nursing_home_count hotel_count theatre_count cinema_count secondary_vocational_education_institutions_count university_count stadium_count emergency_medical_service_stations_count kindergarten_count hostel_count ... post_count swimming_pool_count library_count guest_house_count fire_safety_facilities_count restaurant_count police_count museum_count bank_count pitch_count
14296 0.014379 -0.000832 0.001763 0.016099 0.039391 0.479495 0.105324 0.005666 0.296905 0.010605 ... 0.072575 0.020337 0.040035 0.033931 0.033148 0.092938 0.037970 0.028900 0.064677 0.987647
3310 0.012930 0.803748 0.081753 0.202503 1.084357 0.137536 1.110656 0.230271 5.922528 0.070684 ... 0.302535 0.109361 0.363499 0.212221 0.722693 1.204709 0.836241 0.587259 0.489967 17.555252
16148 0.249490 13.239592 0.481125 1.550649 3.687493 1.599196 2.781085 0.267505 23.173780 0.916293 ... 1.557643 13.684464 1.686338 0.224791 1.782080 15.103293 2.445613 1.884544 1.571543 84.765625
17455 -0.003019 0.009737 -0.004039 0.006393 0.019949 0.001556 0.085916 -0.000057 0.383974 -0.001581 ... 0.144662 0.030829 0.018814 0.171654 0.001377 0.121550 0.095347 0.025819 0.004237 0.626067
16291 0.013056 0.088273 0.002856 0.019799 0.034916 0.024223 0.249959 0.007760 0.771067 0.012497 ... 0.444398 0.005997 0.135191 0.156082 0.167819 0.178831 0.075085 0.083318 0.011640 1.925919

5 rows × 28 columns

Calculating statistics

[9]:

stats_df = sr.calculate_interval_stats(y_pred, pi_lower, pi_upper, y_test)

Showing prediction intervals

[10]:

pi_lower.head()

[10]:
nursing_home_count hotel_count theatre_count cinema_count secondary_vocational_education_institutions_count university_count stadium_count emergency_medical_service_stations_count kindergarten_count hostel_count ... post_count swimming_pool_count library_count guest_house_count fire_safety_facilities_count restaurant_count police_count museum_count bank_count pitch_count
14296 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
3310 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
16148 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
17455 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
16291 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

5 rows × 28 columns

[11]:

pi_upper.head()

[11]:
nursing_home_count hotel_count theatre_count cinema_count secondary_vocational_education_institutions_count university_count stadium_count emergency_medical_service_stations_count kindergarten_count hostel_count ... post_count swimming_pool_count library_count guest_house_count fire_safety_facilities_count restaurant_count police_count museum_count bank_count pitch_count
14296 0.800304 1.000252 0.000069 0.000049 0.158654 0.000215 1.000184 0.000031 1.995936 0.000027 ... 1.833845 0.000129 0.949811 0.106655 1.000122 1.000830 0.134224 0.355738 1.000056 2.654914
3310 0.800304 6.321808 0.934067 1.099999 3.683720 2.214749 4.400317 1.064502 14.263688 1.000000 ... 1.839819 0.999986 1.761400 1.722426 2.808025 8.334574 3.968300 4.091033 2.599319 38.652114
16148 1.200304 49.796808 2.152624 2.460287 6.560348 4.628682 6.023683 1.729038 36.571412 2.704440 ... 2.497692 81.805113 3.051844 11.955599 4.280855 22.667316 5.653427 4.226640 2.697712 103.612669
17455 0.000027 1.000252 0.000069 0.000049 0.000340 0.000215 1.821733 0.000031 1.011686 0.000027 ... 1.523892 0.000129 0.877675 1.335098 1.000122 1.000830 0.436661 0.297617 1.000056 2.118286
16291 0.800304 1.669386 0.459638 0.000049 1.100385 0.000215 1.902446 0.000031 3.415354 0.000027 ... 1.780039 0.999986 1.114143 2.099003 1.669316 1.999971 1.647817 1.395385 1.000056 6.649416

5 rows × 28 columns

Showing validation statistics

[12]:

stats_df

[12]:
y coverage_percentage mean_interval_width mse rmse mae r2
nursing_home_count 99.473684 0.421085 0.024088 0.155203 0.046052 -0.272759
hotel_count 97.368421 3.772173 26.111585 5.109950 1.009036 0.276750
theatre_count 96.842105 0.547877 0.358543 0.598784 0.169514 0.364897
cinema_count 97.192982 0.437983 0.102323 0.319880 0.108495 0.146454
secondary_vocational_education_institutions_count 96.491228 1.577391 1.027018 1.013419 0.379245 0.678409
university_count 96.491228 1.205866 4.898239 2.213197 0.493575 0.370108
stadium_count 97.192982 2.360784 1.096632 1.047202 0.503562 0.576055
emergency_medical_service_stations_count 97.894737 0.231561 0.102410 0.320016 0.096631 0.361666
kindergarten_count 96.315789 8.174934 26.301057 5.128456 1.679345 0.827755
hostel_count 97.192982 0.457503 2.267950 1.505971 0.253160 -1.489247
park_count 95.614035 6.418581 16.817835 4.100955 1.457848 0.624552
multifunctional_center_count 98.596491 2.534885 1.495145 1.222761 0.450774 0.535752
pharmacy_count 96.491228 1.901912 4.224587 2.055380 0.569804 0.435232
sports_halls_count 97.368421 0.367853 0.672834 0.820265 0.181746 -0.186414
hospital_count 98.771930 3.738856 3.686516 1.920030 0.798328 0.515673
school_count 96.315789 8.363565 17.436128 4.175659 1.668863 0.818749
mall_count 97.368421 4.266809 7.596958 2.756258 0.893124 0.747427
polyclinic_count 97.192982 1.962443 1.069309 1.034074 0.393067 0.545571
post_count 97.368421 1.930996 0.866621 0.930925 0.459623 0.535083
swimming_pool_count 97.894737 2.582380 30.738956 5.544272 0.565730 0.465928
library_count 96.842105 1.601277 0.328383 0.573047 0.284873 0.448237
guest_house_count 96.140351 2.934491 12.888878 3.590108 0.845219 0.317789
fire_safety_facilities_count 97.543860 1.807037 0.529556 0.727706 0.368584 0.679995
restaurant_count 95.964912 4.815160 9.405612 3.066857 1.086276 0.558713
police_count 97.192982 1.719624 0.833359 0.912885 0.404326 0.646310
museum_count 95.964912 1.765682 1.956423 1.398722 0.454936 -0.054753
bank_count 98.070175 1.761644 0.774361 0.879978 0.325235 0.448727
pitch_count 95.789474 21.877662 114.181879 10.685592 4.007621 0.905130 
[22]:

import numpy as np
import matplotlib.pyplot as plt

def visualize_prediction(target_name : str, y_pred, pi_lower, pi_upper, y_test):

  # Filter data where y_test values are positive
  mask = y_test[target_name].values > 0
  filtered_pred_df = y_pred[mask]
  filtered_pi_lower = pi_lower[mask]
  filtered_pi_upper = pi_upper[mask]
  filtered_y_test = y_test[mask]

  # Parameters for slicing (default is full filtered range)
  start_idx = 0  # Starting index (can be changed)
  end_idx = 50   # Ending index (can be changed)

  # Extract data from filtered DataFrames with slicing
  x = np.linspace(0, len(filtered_pred_df) - 1, len(filtered_pred_df))[start_idx:end_idx + 1]
  y_pred = filtered_pred_df[target_name].values[start_idx:end_idx + 1]
  y_test = filtered_y_test[target_name].values[start_idx:end_idx + 1]

  # Extract intervals from pi_df (now stored as tuples)
  lower_bound = filtered_pi_lower[target_name].values[start_idx:end_idx + 1]
  upper_bound = filtered_pi_upper[target_name].values[start_idx:end_idx + 1]

  # Create the plot
  plt.figure(figsize=(10, 6))
  plt.plot(x, y_pred, label='Predictions', marker='o')
  plt.plot(x, y_test, label='True Values', marker='s')
  plt.fill_between(x, lower_bound, upper_bound, alpha=0.2, label='Prediction Intervals')

  # Configure the plot
  plt.xlabel('Indices')
  plt.ylabel('Value')
  plt.title('Comparison of Predictions and True Values with Intervals')
  plt.legend()
  plt.grid(True)
  plt.xticks(x)
  plt.tight_layout()

[23]:

target_name = 'pharmacy_count'

visualize_prediction(target_name, y_pred, pi_lower, pi_upper, y_test)
../../_images/examples_machine_learning_social_22_0.png

Error for different targets

[19]:

import matplotlib.pyplot as plt
import numpy as np

# Assuming stats_df is obtained from calculate_interval_stats and has been transposed
# Example: pred_df, pi_df = ir.predict_with_intervals(X_test, inverse_transform=False)
# stats_df = ir.calculate_interval_stats(pred_df, pi_df, y_test).T

# Extract MSE values and target names from transposed stats_df
mse = stats_df['mse'].values
target_names = stats_df.index

# Sort by descending MSE
sorted_indices = np.argsort(mse)[::-1]
sorted_mse = mse[sorted_indices]
sorted_targets = target_names[sorted_indices]

# Create horizontal bar plot
plt.figure(figsize=(16, 30))
plt.barh(sorted_targets, sorted_mse, color='skyblue')

# Add labels and title
plt.ylabel('Target')
plt.xlabel('Error Value')
plt.title('Mean Squared Errors for Different Targets')

# Add grid and save plot
plt.grid(axis='x', linestyle='--', alpha=0.7)
plt.tight_layout()
plt.plot()

[19]:

[]
../../_images/examples_machine_learning_social_24_1.png

LIME Prediction

[ ]:

import pandas as pd
from lime.lime_tabular import LimeTabularExplainer
import matplotlib.pyplot as plt

# Target column
column = 'university_count'

# Fill missing values in y_train
y_train = y_train.fillna(0)

# Use ir model for predictions
pred_df, pi_df = sr.evaluate_with_intervals(X_test, inverse_transform=False)

# Get MSE from stats_df (assuming stats_df is transposed)
stats_df = sr.calculate_interval_stats(pred_df, pi_df, y_test)
mse = stats_df.loc[column, 'mse']

# Print MSE
print(f'Mean Squared Error: {mse:.3f}')

# Helper function for LIME to predict with ir model for the specific column
def predict_fn(X):
    # Convert input to DataFrame with correct column names
    X_df = pd.DataFrame(X, columns=X_train.columns)
    # Get predictions from ir model
    pred_df, _ = sr.evaluate_with_intervals(X_df, inverse_transform=False)
    # Return predictions for the specific column
    return pred_df[column].values

# Create LIME explainer for tabular data
explainer = LimeTabularExplainer(
    training_data=X_train.values,  # Convert DataFrame to numpy array
    feature_names=X_train.columns,  # Feature names
    class_names=[column],  # Target name for regression
    mode='regression',  # Task type: regression
    random_state=42
)

# Select one instance for interpretation (e.g., first from test set)
i = 0
instance = X_test.iloc[i].values.reshape(1, -1)

# Get explanation for the selected instance
explanation = explainer.explain_instance(instance[0], predict_fn)

# Print prediction and real value
print("Predicted value:", pred_df[column].iloc[i])
print("Real value:", y_test[column].iloc[i])
print("Explanation:")
print(explanation.as_list())

# Get explanation as a list
explanation_list = explanation.as_list()

# Split explanation into features and weights
features, weights = zip(*explanation_list)

# Create horizontal bar plot
plt.figure(figsize=(10, 6))
plt.barh(features, weights, color='skyblue')

# Add labels and title
plt.xlabel("Feature Weight")
plt.ylabel("Feature")
plt.title("LIME Explanation for Prediction")

# Add grid and save plot
plt.grid(axis='x', linestyle='--', alpha=0.7)
plt.tight_layout()
plt.plot()

Mean Squared Error: 4.898
Predicted value: 0.479
Real value: 0.0
Explanation:
[('business > 15183.04', 2.9534802613554785), ('population <= 1136.00', -0.5860909439229887), ('transport <= 0.00', 0.3809469174382578), ('74914.77 < industrial <= 528208.61', -0.2871944980298722), ('buildings_count <= 4.00', -0.23936306374227934), ('area <= 63000.00', -0.1874804409055796), ('34922.80 < special <= 156359.65', -0.1421858470028649), ('footprint_area <= 0.00', -0.12259403792263908), ('longitude <= -883053.12', 0.12092895926078623), ('444312.69 < recreation <= 4056059.51', 0.10099483719723343)]

[]
../../_images/examples_machine_learning_social_26_2.png