Python classification.accuracy_score方法代码示例

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def predict(self, model, X, y):
        Y_pred = model.predict_proba_dict(X) 
        df = pd.DataFrame(Y_pred).values
        print('Accuracy: ', accuracy_score(y, np.argmax(df, axis=1)))
        return df 

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def predict(self, model, X, y, ContinuousColumnName, CategoricalColumnName):
        predictions = np.array(list(model.predict_proba(input_fn=lambda: self.input_fn(X, y, ContinuousColumnName, CategoricalColumnName))))
        
        results = model.evaluate(input_fn=lambda: self.input_fn(X, y, ContinuousColumnName, CategoricalColumnName), steps=1)

        for key in sorted(results):
            print("%s: %s"%(key, results[key]))
        print('Accuracy: ', accuracy_score(y, tf.argmax(predictions, axis=1)))
        return predictions 

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def predict(self, model, X, y):
        predictions = model.predict_proba(X)
        if np.isfinite(y).all():
            print('Accuracy: ', accuracy_score(y, np.argmax(predictions, axis=1)))
        return predictions 

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def prediction_score(train_X, train_y, test_X, test_y, metric, model):
    # if the train labels are always the same
    values_train = set(train_y)
    if len(values_train) == 1:
        # predict always that value
        only_value_train = list(values_train)[0]
        test_pred = np.ones_like(test_y) * only_value_train

    # if the train labels have different values
    else:
        # create the model
        if model == "random_forest_classifier":
            m = RandomForestClassifier(n_estimators=10)
        elif model == "logistic_regression":
            m = LogisticRegression()
        else:
            raise Exception("Invalid model name.")

        # fit and predict
        m.fit(train_X, train_y)
        test_pred = m.predict(test_X)

    # calculate the score
    if metric == "f1":
        return f1_score(test_y, test_pred)
    elif metric == "accuracy":
        return accuracy_score(test_y, test_pred)
    else:
        raise Exception("Invalid metric name.") 

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def train_and_eval(ngram_range=(1, 1), max_features=None, max_df=1.0, C=1.0):
    """Train and eval newsgroup classification.

    :param ngram_range: ngram range
    :param max_features: the number of maximum features
    :param max_df: max document frequency ratio
    :param C: Inverse of regularization strength for LogisticRegression
    :return: metrics
    """
    # Loads train and test data.
    train_data = fetch_20newsgroups(subset='train')
    test_data = fetch_20newsgroups(subset='test')

    # Define the pipeline.
    pipeline = Pipeline([
        ('tfidf', TfidfVectorizer()),
        ('clf', LogisticRegression(multi_class='auto'))
    ])

    # Set pipeline parameters.
    params = {
        'tfidf__ngram_range': ngram_range,
        'tfidf__max_features': max_features,
        'tfidf__max_df': max_df,
        'clf__C': C,
    }
    pipeline.set_params(**params)
    print(pipeline.get_params().keys())

    # Train the model.
    pipeline.fit(train_data.data, train_data.target)
    # Predict test data.
    start_time = time()
    predictions = pipeline.predict(test_data.data)
    inference_time = time() - start_time
    avg_inference_time = 1.0 * inference_time / len(test_data.target)
    print("Avg. inference time: {}".format(avg_inference_time))

    # Calculate the metrics.
    accuracy = accuracy_score(test_data.target, predictions)
    recall = recall_score(test_data.target, predictions, average='weighted')
    f1 = f1_score(test_data.target, predictions, average='weighted')
    metrics = {
        'accuracy': accuracy,
        'recall': recall,
        'f1': f1,
    }

    return metrics 

# 需要导入模块: from sklearn.metrics import classification [as 别名]
# 或者: from sklearn.metrics.classification import accuracy_score [as 别名]
def train_and_eval(output, ngram_range=(1, 1), max_features=None, max_df=1.0, C=1.0):
    """Train and eval newsgroup classification.

    :param ngram_range: ngram range
    :param max_features: the number of maximum features
    :param max_df: max document frequency ratio
    :param C: Inverse of regularization strength for LogisticRegression
    :return: metrics
    """
    # Loads train and test data.
    train_data = fetch_20newsgroups(subset='train')
    test_data = fetch_20newsgroups(subset='test')

    # Define the pipeline.
    pipeline = Pipeline([
        ('tfidf', TfidfVectorizer()),
        ('clf', LogisticRegression(multi_class='auto'))
    ])

    # Set pipeline parameters.
    params = {
        'tfidf__ngram_range': ngram_range,
        'tfidf__max_features': max_features,
        'tfidf__max_df': max_df,
        'clf__C': C,
    }
    pipeline.set_params(**params)
    print(pipeline.get_params().keys())

    # Train the model.
    pipeline.fit(train_data.data, train_data.target)
    # Predict test data.
    start_time = time()
    predictions = pipeline.predict(test_data.data)
    inference_time = time() - start_time
    avg_inference_time = 1.0 * inference_time / len(test_data.target)
    print("Avg. inference time: {}".format(avg_inference_time))

    # Calculate the metrics.
    accuracy = accuracy_score(test_data.target, predictions)
    recall = recall_score(test_data.target, predictions, average='weighted')
    f1 = f1_score(test_data.target, predictions, average='weighted')
    metrics = {
        'accuracy': accuracy,
        'recall': recall,
        'f1': f1,
    }

    # Persistent the model.
    joblib.dump(pipeline, output)

    return metrics