[FIXED] Ho can save the loss function in a for cycle?

Issue

Goodmorning, I built neural network in order to predict a physical quantietis. I want to run 10 times the model in order to see the stability of the model. How can I create a DataFrame with all the loss function (both validation and training) evaluated along the 10 try?

for i in range(10): #per in numero di esperimenti

    test_size = 0.2

    dataset = pd.read_csv('CompleteDataSet_original_Clean_TP.csv', decimal=',', delimiter = ";")

    label = dataset.iloc[:,-1]
    features = dataset[feat_labels]

    y_max_pre_normalize = max(label)
    y_min_pre_normalize = min(label)

    def denormalize(y):
        final_value = y*(y_max_pre_normalize-y_min_pre_normalize)+y_min_pre_normalize
        return final_value


    X_train1, X_test1, y_train1, y_test1 = train_test_split(features, label, test_size = test_size, shuffle = True)

    y_test2 = y_test1.to_frame()
    y_train2 = y_train1.to_frame()

    scaler1 = preprocessing.MinMaxScaler()
    scaler2 = preprocessing.MinMaxScaler()
    X_train = scaler1.fit_transform(X_train1)
    X_test = scaler2.fit_transform(X_test1)


    scaler3 = preprocessing.MinMaxScaler()
    scaler4 = preprocessing.MinMaxScaler()
    y_train = scaler3.fit_transform(y_train2)
    y_test = scaler4.fit_transform(y_test2)

    from keras import backend as K


    # =============================================================================
    # Creo la rete
    # =============================================================================
    optimizer = tf.keras.optimizers.Adam(lr=0.001)
    model = Sequential()

    model.add(Dense(100, input_shape = (X_train.shape[1],), activation = 'relu',kernel_initializer='glorot_uniform'))
    model.add(Dropout(0.2))
    model.add(Dense(100, activation = 'relu',kernel_initializer='glorot_uniform'))
    model.add(Dropout(0.2))
    model.add(Dense(100, activation = 'relu',kernel_initializer='glorot_uniform'))
    model.add(Dropout(0.2))
    model.add(Dense(100, activation = 'relu',kernel_initializer='glorot_uniform'))

    model.add(Dense(1,activation = 'linear',kernel_initializer='glorot_uniform'))

    model.compile(loss = 'mse', optimizer = optimizer, metrics = ['mse', r2_score])

    history = model.fit(X_train, y_train, epochs = 200,
                        validation_split = 0.1, shuffle=False, batch_size=250)


    history_dict = history.history

    loss_values = history_dict['loss']
    val_loss_values = history_dict['val_loss']

    y_train_pred = model.predict(X_train)
    y_test_pred = model.predict(X_test)

    y_train_pred = denormalize(y_train_pred)
    y_test_pred = denormalize(y_test_pred)
    from sklearn.metrics import r2_score


    from sklearn import metrics

    r2_test.append(r2_score(y_test_pred, y_test1))
    r2_train.append(r2_score(y_train_pred, y_train1))


     # Measure MSE error.  
    MSE_test.append(metrics.mean_squared_error(y_test_pred,y_test1))
    MSE_train.append(metrics.mean_squared_error(y_train_pred,y_train1))
    RMSE_test.append(np.sqrt(metrics.mean_squared_error(y_test_pred,y_test1)))
    RMSE_train.append(np.sqrt(metrics.mean_squared_error(y_train_pred,y_train1)))

Solution

Without even using Pandas, just append them to a list. You can then use the pd.DataFrame() constructor if you really need a dataframe.

loss_and_val_loss = []
for i in range(...):
    # ...
    loss_values = history_dict['loss']
    val_loss_values = history_dict['val_loss']
    loss_and_val_loss.append((loss_values, val_loss_values))
# ...

Assuming both of those _valueses are lists of numbers, one per epoch, this can be pivoted into dataframes as follows:

# (Example data with two trials, each with 3 epochs)
loss_and_val_loss = [
    ([1, 2, 3], [4, 5, 6]),
    ([7, 8, 9], [10, 11, 12]),
]
losses, val_losses = zip(*loss_and_val_loss)
losses_df = pd.DataFrame(losses)
val_losses_df = pd.DataFrame(val_losses)

Answered By - AKX