I want the solution in Python language

Part 1: a) Check the code from mnist_mlp,py and run it. b) Leave it with only one hidden layer, remove the dropouts and run it again. c) Change the activation of the hidden layer to a "sigmoi d, the loss function to mse (mean squared error). d) Change the optimization algorithm to "sgd" (Stochastic Gradient Descent). Part 2: a) Check the code from mnist cnn.py and run it. b) Add a Max Poo 11D layer after the first convolutional layer, and a dropout of 25% after it. Also, change the optimization algorithm to "adam". Part 3: a) Check the code from lstm. text generation.py and run it. b) Change the length of the inputs to 80 characters. c) Change the length back to its original size and load the previously trained model, with the following syntax: model. load_weights ('model_latm_text.h.') d) Train the same network for 50 epochs with Isaac Asimov's tale "The Last Question", available at the_last_question.txt. For the sake of time, remove the callback parameter from the model.fit (...) function and add a call to on epoch end (0,0) at the end of the program, so that you can check text generated by the model. Part 4: a) Check the code from data_handler.py and convlstm. py and run it. b) Change the dropout rate on the LSTM layer to 80%. c) Train the network with only the 6 first features (columns) of the data