progress bar for GridSearchCV

Question

I am building CNN algorithm that will output some values. I am using GridSearchCV for parameters tuning and I want to implement progress bar for handling with large datasets but I do not know how to.

Input:

def create_model(units):
    model = Sequential() # defining model
    model.add(Conv1D(filters = 64, kernel_size = 5, activation = 'relu', input_shape = (max_sensor_length, SENSORS))) # first convolution layer
    model.add(MaxPool1D(pool_size = 8)) # first pooling layer
    model.add(Conv1D(filters = 16, kernel_size = 5, activation = 'relu')) # 2. layer
    model.add(MaxPool1D(pool_size = 4))
    model.add(Conv1D(filters = 16, kernel_size = 5, activation = 'relu')) # 3. layer
    model.add(MaxPool1D(pool_size = 4))
    model.add(Flatten()) # adding fully connected layer
    model.add(Dense(units = units, activation = 'relu'))
    model.add(Dense(y_ohe.shape[1], activation = 'softmax')) # adding output layer
    model.compile(loss = 'categorical_crossentropy', 
                  optimizer = 'adam', 
                  metrics = ['accuracy'], 
                  ) # compiling the model
    return model

model = KerasRegressor(model = create_model, units = 5)

batch_size = [5, 10, 30, 50, 100]
epochs = [5, 10, 30, 50, 100]
units = [5, 10, 15, 30, 50, 100]

param_grid = dict(batch_size = batch_size, 
                    epochs = epochs,
                    units = units)

grid = GridSearchCV(estimator = model, 
                    param_grid = param_grid, 
                    n_jobs = -1, 
                    cv = 3, 
                    verbose = 10
                    )

grid_result = grid.fit(X_train, 
                       Y_train, 
                       validation_data = (X_test, Y_test)
                       )

Answer 1

From the sklearn documentation on gridsearchCV

verbose (int) Controls the verbosity: the higher, the more messages.

1 : the computation time for each fold and parameter candidate is displayed;

2 : the score is also displayed;

3 : the fold and candidate parameter indexes are also displayed together with the starting time of the computation.

Just set verbose = 1 in the gridsearch parameters (not sure if 10 works?)

Probably you will regret this later as you will end up with a wall of text.

If you are talking about a progress bar for the CNN I would look at tqdm

Answer 2

Use this function, and instead of doing grid.fit(X_train, Y_train, validation_data=(X_test,Y_test)), you simply do fit(grid, X_train, Y_train, validation_data=(X_test,Y_test))

import sys
from tqdm.notebook import tqdm
# from tqdm import tqdm if not in notebook

def fit(model, *args, **kwargs):
    class BarStdout:
        def write(self, text):
            if "totalling" in text and "fits" in text:
                self.bar_size = int(text.split("totalling")[1].split("fits")[0][1:-1])
                self.bar = tqdm(range(self.bar_size))
                self.count = 0
                return
            if "CV" in text and hasattr(self,"bar"):
                self.count += 1
                self.bar.update(n=self.count-self.bar.n)
                if self.count%(self.bar_size//10)==0:
                    time.sleep(0.1)
        def flush(self, text=None):
            pass
    default_stdout= sys.stdout
    sys.stdout = BarStdout()
    model.verbose = 10
    model.fit(*args, **kwargs)
    sys.stdout = default_stdout
    return model