Python: Neural Network - TypeError: 'History' object is not subscriptable

Question

I have been practicing building and comparing neural networks using Keras and Tensorflow in python, but when I look to plot the models for comparisons I am receiving an error:

TypeError: 'History' object is not subscriptable 

Here is my code for the three models:

############################## Initiate model 1 ############################### # Model 1 has no hidden layers from keras.models import Sequential model1 = Sequential()  # Get layers from keras.layers import Dense # Add first layer n_cols = len(X.columns) model1.add(Dense(units=n_cols, activation='relu', input_shape=(n_cols,))) # Add output layer model1.add(Dense(units=2, activation='softmax'))  # Compile the model model1.compile(loss='categorical_crossentropy', optimizer='adam', metrics=  ['accuracy'])   # Define early_stopping_monitor from keras.callbacks import EarlyStopping early_stopping_monitor = EarlyStopping(patience=2)  # Fit model model1.fit(X, y, validation_split=0.33, epochs=30, callbacks=  [early_stopping_monitor], verbose=False)   ############################## Initiate model 2 ############################### # Model 2 has 1 hidden layer that has the mean number of nodes of input and output layer model2 = Sequential()  # Add first layer model2.add(Dense(units=n_cols, activation='relu', input_shape=(n_cols,))) # Add hidden layer import math model2.add(Dense(units=math.ceil((n_cols+2)/2), activation='relu')) # Add output layer model2.add(Dense(units=2, activation='softmax'))  # Compile the model model2.compile(loss='categorical_crossentropy', optimizer='adam', metrics=  ['accuracy'])   # Fit model model2.fit(X, y, validation_split=0.33, epochs=30, callbacks=  [early_stopping_monitor], verbose=False)  ############################## Initiate model 3 ############################### # Model 3 has 1 hidden layer that is 2/3 the size of the input layer plus the size of the output layer model3 = Sequential()  # Add first layer model3.add(Dense(units=n_cols, activation='relu', input_shape=(n_cols,))) # Add hidden layer model3.add(Dense(units=math.ceil((n_cols*(2/3))+2), activation='relu')) # Add output layer model3.add(Dense(units=2, activation='softmax'))  # Compile the model model3.compile(loss='categorical_crossentropy', optimizer='adam', metrics=  ['accuracy'])   # Fit model model3.fit(X, y, validation_split=0.33, epochs=30, callbacks=  [early_stopping_monitor], verbose=False)   # Plot the models plt.plot(model1.history['val_loss'], 'r', model2.history['val_loss'], 'b',  model3.history['val_loss'], 'g') plt.xlabel('Epochs') plt.ylabel('Validation score') plt.show() 

I have no problems with running any of my models, getting predicted probabilities, plotting ROC curves, or plotting PR curves. However, when I attempt to plot the three curves together I am getting an error from this area of my code:

model1.history['val_loss']  TypeError: 'History' object is not subscriptable 

Does anyone have experience with this type of error and can lead me to what I am doing wrong?

Thank you in advance.

Answer 1

Call to model.fit() returns a History object that has a member history, which is of type dict.

So you can replace :

model2.fit(X, y, validation_split=0.33, epochs=30, callbacks=  [early_stopping_monitor], verbose=False) 

with

history2 = model2.fit(X, y, validation_split=0.33, epochs=30, callbacks=  [early_stopping_monitor], verbose=False) 

Similarly for other models.

and then you can use :

plt.plot(history1.history['val_loss'], 'r', history2.history['val_loss'], 'b',  history3.history['val_loss'], 'g') 

Answer 2

The accepted answer is great. However, in case anyone is trying to access history without storing it during fit, try the following:

Since val_loss is not an attribute on the History object and not a key that you can index with, the way you wrote it won't work. However, what you can try is to access the attribute history in the History object, which is a dict that should contain val_loss as a key.

so, replace:

plt.plot(model1.history['val_loss'], 'r', model2.history['val_loss'], 'b',  model3.history['val_loss'], 'g') 

with

plt.plot(model1.history.history['val_loss'], 'r', model2.history.history['val_loss'], 'b',  model3.history.history['val_loss'], 'g')