def grid_search(path_to_texts, param_grid, word_to_check) :
params=ParameterGrid(param_grid)
results=defaultdict(list)
for ind in range(params.__len__()) :
print('-'*100)
print('Done {}/{}'.format(ind,params.__len__()))
curr_params=params.__getitem__(ind)
print('\tCurrent set of params : {}'.format(curr_params))
model=fasttext.train_unsupervised(path_to_texts, dim=300,model='skipgram',epoch=curr_params['epoch'],ws=curr_params['ws'],minCount=curr_params['minCount'],thread=4,bucket=100000,verbose=0)
results['ws'].append(curr_params['ws'])
results['epoch'].append(curr_params['epoch'])
results['minCount'].append(curr_params['minCount'])
z=get_cos_sim_from_model(word_to_check,model)
results['cos_sim'].append(z)
print('\tNumber of words in vocab : {}'.format(len(model.words)))
print('\tMost similar to "{}" : '.format(word_to_check))
for key,value in z.items() :
print('\t',key,' : ', value)
del model
return results
if minimum == maximum:
alle[feature] = [minimum] * 25
else:
diff = (maximum - minimum) / 25
alle[feature] = np.arange(minimum, maximum, diff)
else:
alle[feature] = X[feature].unique()
from sklearn.model_selection import ParameterGrid
grid = ParameterGrid(alle)
# Evaluate the Grid
grid_results = pd.DataFrame()
n_samples = 50000
iteration = 0
while iteration < n_samples:
params = grid[randint(0, grid.__len__())]
params = check_conditions(params)
test = np.array(pd.Series(params)).reshape(1, -1)
params['prediction'] = full_emp_randomForest.predict(test)[0]
grid_results = grid_results.append(params, ignore_index=True)
iteration += 1
# Train Gaussian Process with results
hyperparameter = grid_results[choice]
prediction = grid_results['prediction']
standardScaler = StandardScaler()
hyperparameter = standardScaler.fit_transform(
hyperparameter.values.reshape(-1, 1))
rbf = 1.0**2 * RBF(length_scale=5.0, length_scale_bounds=(1e-1, 20.0))
gaussianProcessRegressor = GaussianProcessRegressor(kernel=rbf)
batch_offset = 0
