ececet's picture
.
780bb93
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
from sklearn.metrics import mean_absolute_error
sns.set_theme()
def read_results(filename):
with open(filename, "r") as f:
lines = f.readlines()
preds_values = []
actual_values = []
mae_values = []
for line in lines:
if line.startswith("Preds:"):
preds = line.replace("[", "")
preds = preds.replace("]", "")
preds = preds.strip("Preds:")
preds = preds.strip()
preds = preds.split(",")
preds = [p.strip() for p in preds]
preds = np.asarray([float(p) for p in preds])
preds_values.append(preds)
if line.startswith("Actual:"):
actual = line.replace("[", "")
actual = actual.replace("]", "")
actual = actual.strip("Actual values:")
actual = actual.strip()
actual = actual.split(",")
actual = [a.strip() for a in actual]
actual = np.asarray([float(a) for a in actual])
actual_values.append(actual)
if line.startswith("MAE"):
mae = float(line.split()[-1])
mae_values.append(mae)
return preds_values, actual_values, mae_values
def plot_distribution(preds_values, actual_values, mae_values, model_name, threshold, oversampled):
for i in range(2):
if i == 0:
input_type = "BoW"
else:
input_type = "TF-IDF"
preds = preds_values[i]
actual = actual_values[i]
mae = mae_values[i]
res = pd.DataFrame()
res["Prediction"] = preds
res["Actual"] = actual
sns.displot(res, kind="kde")
plt.xlabel("Home standard score")
plt.title(f"Model: {model_name}, Input type: {input_type}, MAE: {mae}, Threshold:{threshold}",
fontsize = 10)
plt.ylim(-0.03, 2.5)
plt.tight_layout()
plt.savefig(f"figs/{model_name}_{input_type}_{threshold[0]}_{threshold[1]}_{oversampled}.png")
plt.close()
def print_category_errors(actual, preds):
for i in range(2):
if i == 0:
input_type = "BoW"
else:
input_type = "TF-IDF"
preds = preds_values[i]
actual = actual_values[i]
mae = mae_values[i]
print(input_type)
actual1 = list(actual[np.where(actual < 0.98)])
preds1 = list(preds[np.where(actual < 0.98)])
print(f"Category 1 MAE: {mean_absolute_error(actual1, preds1):.4f}")
print(f"Category 1 correlation: {np.corrcoef(actual1, preds1)[0][1]:.4f}")
print()
actual2 = list(actual[np.where((actual >= 0.98) & (actual < 1.5))])
preds2 = list(preds[np.where((actual >= 0.98) & (actual < 1.5))])
print(f"Category 2 MAE: {mean_absolute_error(actual2, preds2):.4f}")
print(f"Category 2 correlation: {np.corrcoef(actual2, preds2)[0][1]:.4f}")
print()
actual3 = list(actual[np.where((actual >= 1.5) & (actual < 2))])
preds3 = list(preds[np.where((actual >= 1.5) & (actual < 2))])
print(f"Category 3 MAE: {mean_absolute_error(actual3, preds3):.4f}")
print(f"Category 3 correlation: {np.corrcoef(actual3, preds3)[0][1]:.4f}")
print()
actual4 = list(actual[np.where(actual >= 2)])
preds4 = list(preds[np.where(actual >= 2)])
print(f"Category 4 MAE: {mean_absolute_error(actual4, preds4):.4f}")
print(f"Category 4 correlation: {np.corrcoef(actual4, preds4)[0][1]:.4f}")
print()
print(f"Overall corr: {np.corrcoef(actual, preds)[0][1]:.4f}")
if __name__ == "__main__":
filename = "linear_models/lasso_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Lasso", [0.01, 0.99], False)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/lin_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Linear regression", [0.01, 0.99], False)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/sgd_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "SGD Regressor", [0.01, 0.99], False)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "oversampled_False_catboost_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "CatBoostRegressor", [0.01, 0.99], False)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/lasso_0.2_0.8.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Lasso", [0.2, 0.8], False)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/oversampled_lin_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Linear regression", [0.01, 0.99], True)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/oversampled_lasso_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Lasso", [0.01, 0.99], True)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/oversampled_sgd_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "SGD Regressor", [0.01, 0.99], True)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "oversampled_True_catboost_reg_0.01_0.99.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "CatBoostRegressor", [0.01, 0.99], True)
print_category_errors(actual_values, preds_values)
print("============================")
filename = "linear_models/oversampled_lasso_0.15_0.85.txt"
print(filename)
preds_values, actual_values, mae_values = read_results(filename)
#plot_distribution(preds_values, actual_values, mae_values, "Lasso", [0.15, 0.85], True)
print_category_errors(actual_values, preds_values)
print("============================")