public_tests.py

import numpy as np
import pandas as pd
import seaborn as sns
from test_utils import *
from keras import backend
from test_case import *

### ex1
def check_for_leakage_test(target):
    df1 = pd.DataFrame({'patient_id': [0, 1, 2]})
    df2 = pd.DataFrame({'patient_id': [2, 3, 4]})
    expected_output_1 = True
    
    print("Test Case 1\n")
    print("df1")
    print(df1)
    print("df2")
    print(df2)
    print("leakage output:", target(df1, df2, 'patient_id'), "\n-------------------------------------")
    
    df3 = pd.DataFrame({'patient_id': [0, 1, 2]})
    df4 = pd.DataFrame({'patient_id': [3, 4, 5]})
    expected_output_2 = False
    
    print("Test Case 2\n")
    print("df1") ### same heading for df3
    print(df3)
    print("df2") ### same heading for df4
    print(df4)
    print("leakage output:", target(df3, df4, 'patient_id'), "\n")
    
    test_cases = [
        {
            "name":"datatype_check",
            "input": [df1, df2, 'patient_id'],
            "expected": expected_output_1,
            "error":"Data-type mismatch, make sure you are using pandas functions"
        },
        {
            "name":"datatype_check",
            "input": [df3, df4, 'patient_id'],
            "expected": expected_output_2,
            "error":"Datatype mismatch, make sure you are using pandas functions"
        },
        {
            "name": "shape_check",
            "input": [df1, df2, 'patient_id'],
            "expected": expected_output_1,
            "error": "Wrong shape, make sure you are using pandas functions"
        },
        {
            "name": "shape_check",
            "input": [df3, df4, 'patient_id'],
            "expected": expected_output_2,
            "error": "Wrong shape, make sure you are using pandas functions"
        },
        {
            "name": "equation_output_check",
            "input": [df1, df2, 'patient_id'],
            "expected": expected_output_1,
            "error": "Wrong output, make sure you are using pandas functions"
        },
        {
            "name": "equation_output_check",
            "input": [df3, df4, 'patient_id'],
            "expected": expected_output_2,
            "error": "Wrong output, make sure you are using pandas functions"
        }
    ]

    multiple_test(test_cases, target)
    
### ex2
def compute_class_freqs_test(target):
    labels_matrix = np.array(
        [[1, 0, 0],
         [0, 1, 1],
         [1, 0, 1],
         [1, 1, 1],
         [1, 0, 1]]
    )

    print("Labels:")
    print(labels_matrix)
    pos_freqs, neg_freqs = target(labels_matrix)
    print("\nPos Freqs: ", pos_freqs)
    print("Neg Freqs: ", neg_freqs, "\n")
    
    expected_freqs = (np.array([0.8, 0.4, 0.8]), np.array([0.2, 0.6, 0.2]))
    
    test_cases = [
        {
            "name":"datatype_check",
            "input": [labels_matrix],
            "expected": expected_freqs,
            "error": "Data-type mismatch."
        },
        {
            "name": "shape_check",
            "input": [labels_matrix],
            "expected": expected_freqs,
            "error": "Wrong shape."
        },
        {
            "name": "equation_output_check",
            "input": [labels_matrix],
            "expected": expected_freqs,
            "error": "Wrong output."
        }
    ]
    
    multiple_test(test_cases, target)
    
### ex3
def get_weighted_loss_test(target, epsilon, sess):
    y_true, w_p, w_n, y_pred_1, y_pred_2 = get_weighted_loss_test_case(sess)
    
    print("y_true:")
    print(y_true)
    print("\nw_p:")
    print(w_p)
    print("\nw_n:")
    print(w_n)
    print("\ny_pred_1:")
    print(y_pred_1)
    print("\ny_pred_2:")
    print(y_pred_2)
    
    L = target(w_p, w_n, epsilon)
    L1 = L(y_true, y_pred_1).eval(session=sess)
    L2 = L(y_true, y_pred_2).eval(session=sess)
    
    print("\nIf you weighted them correctly, you'd expect the two losses to be the same.")
    print("With epsilon = 1, your losses should be, L(y_pred_1) = -0.4956203 and L(y_pred_2) = -0.4956203\n")
    print("Your outputs:\n")
    print("L(y_pred_1) = ", L1)
    print("L(y_pred_2) = ", L2)
    print("Difference: L(y_pred_1) - L(y_pred_2) = ", L1-L2, "\n")
    
    expected_output_1 = np.float32(-0.4956203)
    expected_output_2 = np.float32(-0.4956203)
    
    test_cases = [
        {
            "name":"datatype_check",
            "input": [y_true, y_pred_1],
            "expected": expected_output_1,
            "error": "Data-type mismatch. Make sure it is a np.float32 value."
        },
        {
            "name":"datatype_check",
            "input": [y_true, y_pred_2],
            "expected": expected_output_2,
            "error": "Data-type mismatch. Make sure it is a np.float32 value."
        },
        {
            "name": "shape_check",
            "input": [y_true, y_pred_1],
            "expected": expected_output_1,
            "error": "Wrong shape."
        },
        {
            "name": "shape_check",
            "input": [y_true, y_pred_2],
            "expected": expected_output_2,
            "error": "Wrong shape."
        },
        {
            "name": "equation_output_check",
            "input": [y_true, y_pred_1],
            "expected": expected_output_1,
            "error": "Wrong output. One possible mistake, your epsilon is not equal to 1."
        },
        {
            "name": "equation_output_check",
            "input": [y_true, y_pred_2],
            "expected": expected_output_2,
            "error": "Wrong output. One possible mistake, your epsilon is not equal to 1."
        }
    ]
    
    multiple_test_weight_loss(test_cases, L, sess)