def deserialize_mlp_regressor(model_dict):
model = MLPRegressor(**model_dict['params'])
model.coefs_ = model_dict['coefs_']
model.loss_ = model_dict['loss_']
model.intercepts_ = model_dict['intercepts_']
model.n_iter_ = model_dict['n_iter_']
model.n_layers_ = model_dict['n_layers_']
model.n_outputs_ = model_dict['n_outputs_']
model.out_activation_ = model_dict['out_activation_']
return model
INDEL_RANGE: List[int] = list(range(MIN_INDEL, MAX_INDEL + 1))
SEQUENCE_MAPPER: Callable[[str], Any] = sequence_to_ordinals
RANDOM_STATE: int = 42
# Model with method 'fit(X, y)'
Model = NamedTuple('Model', [('name', str), ('model', Any), ('shorthand', str),
('preprocess_X', Optional[Callable])])
mlp = MLPRegressor(hidden_layer_sizes=(100, 100, 100, 100, 100, 100),
solver='lbfgs',
max_iter=50,
verbose=True)
mlp.n_layers_ = 8
# feature_map_nystroem = Nystroem(gamma=.2, random_state=1, n_components=10000)
Model = NamedTuple('Model', [('name', str), ('model', Any),
('shorthand', str)])
asdf: Dict[str, Model] = {}
MODELS: Dict[str, Model] = {
'RandomForest':
Model(name='Random Forest',
model=RandomForestRegressor(n_jobs=-1,
bootstrap=True,
random_state=RANDOM_STATE,
verbose=1),
shorthand='rf',
#%%
reg.fit(train, train)
#%% Plot result on training data
output_eval_train = reg.predict(train)
fig = px.scatter(x=train[:,0], y=train[:,1],
labels={'x': 'x[0]', 'y': 'x[1]'})
fig.add_scatter(x=output_eval_train[:,0], y=output_eval_train[:,1], mode='markers')
#%% Plot result on test data
output_eval_test = reg.predict(test)
fig = px.scatter(x=test[:,0], y=test[:,1],
labels={'x': 'x[0]', 'y': 'x[1]'})
fig.add_scatter(x=output_eval_test[:,0], y=output_eval_test[:,1], mode='markers')
# %% Cut the network in half at the hidden 1D layer
# to get values of the hidden parameter
reg.n_layers_ = ((reg.n_layers_ - 2)+1) // 2 + 1
#%%
ae_parm = reg.predict(train)
fig = go.Figure()
fig.add_scatter(x=ae_parm, y=train[:,0],
mode='markers', name='x[0]')
fig.add_scatter(x=ae_parm, y=output_eval_train[:,0],
mode='markers', name='x[0] reduced')
fig.add_scatter(x=ae_parm, y=train[:,1],
mode='markers', name='x[1]')
fig.add_scatter(x=ae_parm, y=output_eval_train[:,1],
mode='markers', name='x[1] reduced')
fig.update_layout(
xaxis_title = 't (hidden curve parameter)',
yaxis_title = 'x[0], x[1]')
# %%
print 'R2 score:', score print 'MAE:', mean_absolute_error(testing_labels,preds), '\n' # PCA + Gradient Boosting Regression gbr = GradientBoostingRegressor(min_samples_split=4) gbr.fit(reduced_training_features, training_labels) preds = gbr.predict(reduced_testing_features) score = gbr.score(reduced_testing_features,testing_labels) print 'PCA + Gradient Boosting Regression Results:' print 'R2 score:', score print 'MAE:', mean_absolute_error(testing_labels,preds) # Multi-Layer Perceptron Regression from sklearn.neural_network import MLPRegressor mlp = MLPRegressor(max_iter=2500) mlp.n_layers_=75 mlp.fit(training_features, training_labels) preds = mlp.predict(testing_features) score = mlp.score(testing_features,testing_labels) print 'Multi-Layer Perceptron Regression Results:' print 'R2 score:', score print 'MAE:', mean_absolute_error(testing_labels,preds), '\n' # PCA + Multi-Layer Perceptron Regression mlp = MLPRegressor(max_iter=2500) mlp.n_layers_=75 mlp.fit(reduced_training_features, training_labels) preds = mlp.predict(reduced_testing_features) score = mlp.score(reduced_testing_features,testing_labels) print 'PCA + Multi-Layer Perceptron Regression Results:' print 'R2 score:', score
list([[7.61211153e-01], [5.29023058e-01], [-6.76783513e-01],
[-1.23527535e-01], [1.04599422e-01], [1.06178562e+00],
[-1.09977597e-43], [-1.22990539e-90], [-3.14851814e-21],
[7.33380751e-01]])
]
net.intercepts_ = [
list([
1.44847648, 1.47542637, 0.51003163, 0.45278632, -0.0056204, 1.53020242,
-0.23453891, -0.00187764, -0.21982535, 1.69397764
]),
list([1.9355952])
]
net.n_outputs_ = 1
net.n_layers_ = 3
net.out_activation_ = "identity"
def dotProd(a, b):
return sum(a[i] * b[i] for i in xrange(len(a)))
def arrToTuple(arr):
tupArr = [tuple(elem) for elem in arr]
return tuple(tupArr)
def isEndState(state):
return state[1] == 0
