Skip to content
Snippets Groups Projects
Commit e70443ac authored by thecml's avatar thecml
Browse files

enabled lime to explain embeddings

parent 80b08d82
No related branches found
No related tags found
No related merge requests found
Pipeline #79270 passed
Stage: test
    Expand all Hide whitespace changes
    Inline Side-by-side
    ml/notebooks/MLP_emb_lime.ipynb
    + 249
    − 222
    View file @ e70443ac
    This diff is collapsed.
    ml/notebooks/MLP_lime.ipynb
    + 24
    − 7
    View file @ e70443ac
    %% Cell type:code id: tags:
    ```
    import tensorflow as tf
    import numpy as np
    import pandas as pd
    from pathlib import Path
    import paths as pt
    import yaml
    from tools import data_loader, preprocessor
    import matplotlib.pyplot as plt
    from sklearn.model_selection import train_test_split
    from sklearn.preprocessing import StandardScaler
    def make_model(input_dim):
    model = tf.keras.models.Sequential()
    model.add(tf.keras.layers.Dense(80,
    input_dim=input_dim,
    activation='relu'))
    model.add(tf.keras.layers.Dropout(0.35))
    model.add(tf.keras.layers.Dense(20, activation='relu'))
    model.add(tf.keras.layers.Dropout(0.15))
    model.add(tf.keras.layers.Dense(10, activation='relu'))
    model.add(tf.keras.layers.Dropout(0.15))
    model.add(tf.keras.layers.Dense(1, activation='sigmoid'))
    metrics = [
    tf.keras.metrics.BinaryAccuracy(name='accuracy'),
    tf.keras.metrics.Precision(name='precision'),
    tf.keras.metrics.Recall(name='recall'),
    tf.keras.metrics.AUC(name='auc'),
    ]
    model.compile(loss='binary_crossentropy',
    optimizer="Adam",
    metrics=metrics)
    return model
    # Load settings
    with open(Path.joinpath(pt.CONFIGS_DIR, "settings.yaml"), 'r') as stream:
    with open(Path.joinpath(pt.CONFIGS_DIR, "complete_emb.yaml"), 'r') as stream:
    settings = yaml.safe_load(stream)
    # Load the data
    file_name = "complete_emb.csv"
    dl = data_loader.CompleteDataLoader(file_name, settings).load_data()
    features = dl.get_features()
    X, y = dl.get_data()
    # Calculate class weight
    neg, pos = np.bincount(y)
    class_weight = preprocessor.get_class_weight(neg, pos)
    # Split data
    X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7,
    stratify=y, random_state=0)
    # Scale the data
    X_train = X_train.reset_index(drop=True)
    X_test = X_test.reset_index(drop=True)
    numeric_feats = ['BirthYear', 'Cluster', 'LoanPeriod', 'NumberAts']
    scaler = StandardScaler()
    X_train_enc = pd.DataFrame(scaler.fit_transform(X_train[numeric_feats]),
    columns=numeric_feats)
    X_test_enc = pd.DataFrame(scaler.transform(X_test[numeric_feats]),
    columns=numeric_feats)
    num_X_train = X_train.drop(numeric_feats, axis=1)
    num_X_test = X_test.drop(numeric_feats, axis=1)
    X_train_sc = pd.concat([num_X_train[['Gender_Male', 'Gender_Female']], X_train_enc,
    num_X_train[['Rand', '1Ats', '2Ats', '3Ats', '4Ats', '5Ats',
    '6Ats', '7Ats', '8Ats', '9Ats', '10Ats']]], axis=1)
    X_test_sc = pd.concat([num_X_test[['Gender_Male', 'Gender_Female']], X_test_enc,
    num_X_test[['Rand', '1Ats', '2Ats', '3Ats', '4Ats', '5Ats',
    '6Ats', '7Ats', '8Ats', '9Ats', '10Ats']]], axis=1)
    ```
    %% Cell type:code id: tags:
    ```
    X_train = np.array(X_train)
    X_train_sc = np.array(X_train_sc)
    X_test = np.array(X_test)
    X_test_sc = np.array(X_test_sc)
    y_train = np.array(y_train)
    y_test = np.array(y_test)
    # Train the model
    model = make_model(input_dim=X_train_sc.shape[1])
    history = model.fit(X_train_sc, y_train, epochs=10,
    class_weight=class_weight,
    batch_size=32, verbose=False)
    batch_size=32, verbose=True)
    ```
    %% Output
    WARNING:tensorflow:From C:\Users\cml\miniconda3\envs\py38-air\lib\site-packages\tensorflow\python\ops\array_ops.py:5043: calling gather (from tensorflow.python.ops.array_ops) with validate_indices is deprecated and will be removed in a future version.
    Instructions for updating:
    The `validate_indices` argument has no effect. Indices are always validated on CPU and never validated on GPU.
    Epoch 1/10
    47/47 [==============================] - 1s 7ms/step - loss: 0.6353 - accuracy: 0.6820 - precision: 0.7183 - recall: 0.9185 - auc: 0.5047
    Epoch 2/10
    47/47 [==============================] - 0s 7ms/step - loss: 0.6051 - accuracy: 0.7153 - precision: 0.7196 - recall: 0.9907 - auc: 0.5196
    Epoch 3/10
    47/47 [==============================] - 0s 6ms/step - loss: 0.5968 - accuracy: 0.7187 - precision: 0.7196 - recall: 0.9981 - auc: 0.5452
    Epoch 4/10
    47/47 [==============================] - 0s 6ms/step - loss: 0.5980 - accuracy: 0.7207 - precision: 0.7208 - recall: 0.9991 - auc: 0.5367
    Epoch 5/10
    47/47 [==============================] - 0s 6ms/step - loss: 0.5860 - accuracy: 0.7180 - precision: 0.7194 - recall: 0.9972 - auc: 0.5813
    Epoch 6/10
    47/47 [==============================] - 0s 6ms/step - loss: 0.5718 - accuracy: 0.7200 - precision: 0.7203 - recall: 0.9991 - auc: 0.6321
    Epoch 7/10
    47/47 [==============================] - 0s 7ms/step - loss: 0.5773 - accuracy: 0.7193 - precision: 0.7198 - recall: 0.9991 - auc: 0.6152
    Epoch 8/10
    47/47 [==============================] - 0s 7ms/step - loss: 0.5717 - accuracy: 0.7213 - precision: 0.7213 - recall: 0.9991 - auc: 0.6401
    Epoch 9/10
    47/47 [==============================] - 0s 7ms/step - loss: 0.5524 - accuracy: 0.7193 - precision: 0.7216 - recall: 0.9935 - auc: 0.6758
    Epoch 10/10
    47/47 [==============================] - 0s 7ms/step - loss: 0.5559 - accuracy: 0.7287 - precision: 0.7278 - recall: 0.9954 - auc: 0.6710
    %% Cell type:code id: tags:
    ```
    def predict(data):
    global model, scaler
    data[:,2:6] = scaler.transform(data[:,2:6])
    y_pred = model.predict(data).reshape(-1, 1)
    y_pred = (y_pred>0.5)
    print(np.array(list(zip(1-y_pred.reshape(data.shape[0]),y_pred.reshape(data.shape[0])))))
    return np.hstack((1-y_pred,y_pred))
    import lime
    import lime.lime_tabular
    explainer = lime.lime_tabular.LimeTabularExplainer(X_train, mode='classification',
    class_names=['No complete', 'Complete'],
    feature_names=features)
    exp = explainer.explain_instance(X_test[27], predict, num_features=X_train.shape[1])
    ```
    %% Output
    [[0 1]
    [1 0]
    [0 1]
    ...
    [0 1]
    [1 0]
    [0 1]]
    %% Cell type:code id: tags:
    ```
    exp.as_list()
    ```
    %% Output
    [('10Ats > 0.03', 0.20651574242348936),
    ('3Ats <= -0.05', 0.192865743355257),
    ('1Ats > 0.07', 0.18353986581301904),
    ('2Ats > -0.03', -0.18296575285910985),
    ('7Ats <= -0.02', -0.16874453649744492),
    ('9Ats <= -0.01', -0.13263677767888812),
    ('8Ats <= 0.06', -0.1273679310001168),
    ('6Ats > 0.08', 0.11597565308592392),
    ('5Ats <= -0.04', 0.11306311651048453),
    ('Gender_Female <= 0.00', -0.08515210039953246),
    ('8.00 < NumberAts <= 14.00', 0.06138616356339817),
    ('LoanPeriod > 1627.50', -0.05305334898011001),
    ('0.00 < Gender_Male <= 1.00', 0.05105843326784843),
    ('29.00 < BirthYear <= 35.00', 0.04509384864824174),
    ('Rand <= 0.25', 0.03759473681614202),
    ('0.03 < 4Ats <= 0.08', 0.0045078832741941325),
    ('Cluster > 11.00', 2.0051051251007468e-05)]
    %% Cell type:code id: tags:
    ```
    model.predict(np.array([X_test_sc[27],]))
    ```
    %% Output
    array([[0.59483784]], dtype=float32)
    ......
    ml/notebooks/MLP_shap.ipynb
    + 81
    − 50
    View file @ e70443ac
    This diff is collapsed.
    ml/src/model/train_embedded_model.py
    + 9
    − 3
    View file @ e70443ac
    ......@@ -14,6 +14,7 @@ import tensorflow as tf
    CASE = "Complete"
    FILENAME = "complete.csv"
    ATS_RESOLUTION = 10
    class NetworkCategory:
    def __init__(self, alias: str, unique_values: int):
    ......@@ -106,14 +107,15 @@ def build_embedding_network(cat_cols, num_cols):
    tf.keras.metrics.BinaryAccuracy(name='accuracy'),
    tf.keras.metrics.Precision(name='precision'),
    tf.keras.metrics.Recall(name='recall'),
    tf.keras.metrics.AUC(name='auc'),
    tf.keras.metrics.AUC(name='roc_auc'),
    tf.keras.metrics.AUC(name='pr_auc', curve='PR')
    ]
    model = tf.keras.Model(inputs=model_inputs, outputs=output_model)
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=metrics)
    model.compile(loss='binary_crossentropy', optimizer='Adam', metrics=metrics)
    return model
    def main():
    ats_cols = {str(i)+'Ats':str for i in range(1, pt.ATS_RESOLUTION+1)}
    ats_cols = {str(i)+'Ats':str for i in range(1, ATS_RESOLUTION+1)}
    df = file_reader.read_csv(pt.PROCESSED_DATA_DIR,
    FILENAME,
    converters=ats_cols)
    ......@@ -198,5 +200,9 @@ def main():
    print("Final recall: %.5f" % recall_score(y_test, y_scores_new))
    print("Final rocauc: %.5f" % roc_auc_score(y_test, y_pred_final))
    from sklearn.metrics import precision_recall_curve, auc
    precision, recall, _ = precision_recall_curve(y_test, y_pred_final)
    print("Final prauc: %.5f" % auc(recall, precision))
    if __name__ == '__main__':
    main()
    0% or .
    You are about to add 0 people to the discussion. Proceed with caution.
    Finish editing this message first!
    Please register or to comment