def anova1rm(Y, A, SUBJ, equal_var=True, roi=None): ''' One-way repeated-measures ANOVA. :Parameters: - *Y* --- (J x Q) numpy array - *A* --- (J x 1) vector of integer group labels - *SUBJ* --- (J x 1) vector of subject labels - *equal_var* --- If *True*, equal group variance will be assumed :Returns: - F : An **spm1d._spm.SPM_F** instance :Example: >>> Y = np.random.randn(9, 101) >>> A = np.array([1,1,1, 2,2,2, 3,3,3]) >>> SUBJ = np.array([1,2,3, 1,2,3, 1,2,3]) >>> F = spm1d.stats.anova1(Y, A, SUBJ) >>> Fi = F.inference(alpha=0.05) >>> Fi.plot() ''' if not equal_var: raise( NotImplementedError( 'Non-sphericity corrections are not yet implemented. Set "equal_var" to "True" to force an assumption of equal variance.' ) ) design = designs.ANOVA1rm(A, SUBJ) model = models.LinearModel(Y, design.X, roi=roi) if (model.dim == 1) and ( design.check_for_single_responses() ): model.fit( approx_residuals=design.contrasts.C[:3] ) else: model.fit( ) F = aov(model, design.contrasts, design.f_terms)[0] return F
def anova3tworm(Y, A, B, C, SUBJ, equal_var=True, roi=None):
'''
Three-way ANOVA with repeated-measures on two factors.
:Parameters:
- *Y* --- (J x Q) numpy array
- *A* --- (J x 1) vector of integer labels for Factor A
- *B* --- (J x 1) vector of integer labels for Factor B (a repeated-measures factor)
- *C* --- (J x 1) vector of integer labels for Factor C (a repeated-measures factor)
- *SUBJ* --- (J x 1) vector of integer subject labels
- *equal_var* --- If *True*, equal group variance will be assumed
:Returns:
- List of seven **spm1d._spm.SPM_F** instances in the following order:
1. Main effect A
2. Main effect B
3. Main effect C
4. Interaction AB
5. Interaction AC
6. Interaction BC
7. Interaction ABC
:Note:
- Non-sphericity correction not implemented. Equal variance must be assumed by setting "equal_var=True".
'''
if equal_var is not True:
raise( NotImplementedError('Non-sphericity correction not implemented. To continue you must assume equal variance and set "equal_var=True".') )
design = designs.ANOVA3tworm(A, B, C, SUBJ)
model = models.LinearModel(Y, design.X, roi=roi)
if (model.dim == 1) and ( design.check_for_single_responses() ):
model.fit( approx_residuals=design.contrasts.C[:8] )
else:
model.fit( )
F = aov(model, design.contrasts, design.f_terms)
return F
def anova3nested(Y, A, B, C, equal_var=True, roi=None):
'''
Three-way fully nested ANOVA.
:Parameters:
- *Y* --- (J x Q) numpy array
- *A* --- (J x 1) vector of integer labels for Factor A
- *B* --- (J x 1) vector of integer labels for Factor B (nested in A)
- *C* --- (J x 1) vector of integer labels for Factor C (nested in B)
- *equal_var* --- If *True*, equal group variance will be assumed
:Returns:
- List of three **spm1d._spm.SPM_F** instances in the following order:
1. Main effect A
2. Main effect B
3. Main effect C
:Note:
- there are no interaction terms in fully-nested designs.
'''
if equal_var is not True:
raise( NotImplementedError('Non-sphericity correction not implemented. To continue you must assume equal variance and set "equal_var=True".') )
design = designs.ANOVA3nested(A, B, C)
model = models.LinearModel(Y, design.X, roi=roi)
model.fit()
F = aov(model, design.contrasts, design.f_terms)
return F
def anova2(Y, A, B, equal_var=True, roi=None): ''' Two-way ANOVA. :Parameters: - *Y* --- (J x Q) numpy array - *A* --- (J x 1) vector of integer labels for Factor A - *B* --- (J x 1) vector of integer labels for Factor B - *equal_var* --- If *True*, equal group variance will be assumed :Returns: - List of three **spm1d._spm.SPM_F** instances in the following order: 1. Main effect A 2. Main effect B 3. Interaction AB ''' if not equal_var: raise( NotImplementedError( 'Non-sphericity corrections are not yet implemented. Set "equal_var" to "True" to force an assumption of equal variance.' ) ) design = designs.ANOVA2(A, B) model = models.LinearModel(Y, design.X, roi=roi) model.fit() F = aov(model, design.contrasts, design.f_terms) # if not equal_var: # Y,X,r = model.Y, model.X, model.eij # QA,QB,C = design.A.get_Q(), design.B.get_Q(), design.contrasts.C.T # Q = QA + QB # u1,u2 = _reml.estimate_df_anova2(Y, X, r, Q, C) # for ff,u in zip(f,u1): # ff.df = u,u2 return F
def test_linear_model(num_genes, num_mirs, num_max_sites, num_features, maxiter):
# generate random data
np.random.seed(0)
# get a random number of sites per mRNA/miRNA interaction
features = np.zeros([num_genes, num_mirs, num_max_sites, num_features])
for i in range(num_genes):
for j in range(num_mirs):
nsites = np.random.choice(num_max_sites)
features[i,j,:nsites,:] = np.random.rand(nsites, num_features)
mask = ((np.abs(np.sum(features, axis=3))) != 0).astype(int)
true_weights = (np.arange(num_features) + 1.0).reshape([1, 1, 1, -1])
true_weights = (true_weights - np.mean(true_weights)) / np.std(true_weights)
labels = np.sum(np.multiply(np.sum(np.multiply(features, true_weights), axis=3), mask), axis=2)
print(features.shape)
print(mask.shape)
print(labels.shape)
tf.reset_default_graph()
features_tensor = tf.placeholder(tf.float32, shape=[None, None, None, num_features], name='features')
mask_tensor = tf.placeholder(tf.float32, shape=[None, None, None], name='nsites')
labels_tensor = tf.placeholder(tf.float32, shape=[None, None], name='labels')
data = {
'features': features_tensor,
'mask': mask_tensor,
'labels': labels_tensor
}
feed_dict = {
features_tensor: features,
mask_tensor: mask,
labels_tensor: labels
}
model = models.LinearModel(num_features)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
model.fit(sess, data, feed_dict, maxiter)
print('True weight diff: {}'.format(np.sum(np.abs(model.vars_evals['coefs'] - true_weights))))
print('Label r2: {}'.format(model.r2))
def anova1(Y, A=None, equal_var=False, roi=None):
'''
One-way ANOVA.
:Parameters (Option 1):
- *Y* --- A list or tuple of (J x Q) numpy arrays
- *equal_var* --- If *True*, equal group variance will be assumed
:Parameters (Option 2):
- *Y* --- (J x Q) numpy array
- *A* --- (J x 1) vector of integer group labels
- *equal_var* --- If *True*, equal group variance will be assumed
:Returns:
- F : An **spm1d._spm.SPM_F** instance
:Example:
>>> F = spm1d.stats.anova1((Y0,Y1,Y2))
>>> Fi = F.inference(alpha=0.05)
>>> Fi.plot()
'''
if isinstance(Y, (list,tuple)):
_datachecks.check('anova1list', Y)
A = np.hstack([[i]*y.shape[0] for i,y in enumerate(Y)])
Y = np.hstack(Y) if Y[0].ndim==1 else np.vstack(Y)
else:
_datachecks.check('anova1', Y, A)
design = designs.ANOVA1(A)
model = models.LinearModel(Y, design.X, roi=roi)
model.fit()
F = aov(model, design.contrasts, design.f_terms)[0]
if not equal_var:
warnings.warn('\nWARNING: Non-sphericity corrections for one-way ANOVA are currently approximate and have not been verified.\n', UserWarning, stacklevel=2)
Y,X,r = model.Y, model.X, model.eij
Q,C = design.A.get_Q(), design.contrasts.C.T
F.df = _reml.estimate_df_anova1(Y, X, r, Q, C)
return F
def __init__(self,
n_stocks,
feature_generator,
gamma=0.95,
alpha=0.1,
epsilon=1.0,
epsilon_min=0.001,
epsilon_decay=0.995,
momentum=0.9):
self.gamma = gamma # discount rate
self.epsilon = epsilon # exploration rate
self.epsilon_min = epsilon_min
self.epsilon_decay = epsilon_decay
self.feature_generator = feature_generator
self.action_mapping, self.action_reverse_mapping = get_all_actions(
n_stocks)
self.n_actions = len(self.action_mapping)
self.model = models.LinearModel(self.feature_generator.size,
self.n_actions,
alpha=alpha,
momentum=momentum)
def anova2rm(Y, A, B, SUBJ, equal_var=True, roi=None):
'''
Two-way repeated-measures ANOVA.
:Parameters:
- *Y* --- (J x Q) numpy array
- *A* --- (J x 1) vector of integer labels for Factor A
- *B* --- (J x 1) vector of integer labels for Factor B
- *SUBJ* --- (J x 1) vector of integer subject labels
- *equal_var* --- If *True*, equal group variance will be assumed
:Returns:
- List of three **spm1d._spm.SPM_F** instances in the following order:
1. Main effect A
2. Main effect B
3. Interaction AB
:Note:
- Non-sphericity correction not implemented. Equal variance must be assumed by setting "equal_var=True".
'''
if equal_var is not True:
raise( NotImplementedError('Non-sphericity correction not implemented. To continue you must assume equal variance and set "equal_var=True".') )
design = designs.ANOVA2rm(A, B, SUBJ)
model = models.LinearModel(Y, design.X, roi=roi)
if (model.dim == 1) and ( design.check_for_single_responses() ):
model.fit( approx_residuals=design.contrasts.C[:5] )
else:
model.fit( )
F = aov(model, design.contrasts, design.f_terms)
# if not equal_var:
# Y,X,r = solver.Y, solver.X, solver.eij
# QA,QB,C = design.A.get_Q(), design.B.get_Q(), [c.C.T for c in design.contrasts]
# Q = QA + QB
# u1,u2 = _reml.estimate_df_anova2(Y, X, r, Q, C)
# for ff,u in zip(f,u1):
# ff.df = u,u2
return F
print("You failed to provide a configuration file. Usage: {0} config.json".
format(sys.argv[0]))
if '.json' not in sys.argv[1]:
raise ValueError("The configuration file must be in JSON format.")
# parse config
config = json.load(open(config_fp, 'r'))
# parse config
config = ConfigParser().parse_config(config=config, mode='evaluate')
# load model
if config['model'] == 'linear':
model = models.LinearModel(input_shape=(config['history_length'], ),
nb_output_units=1,
nb_hidden_units=config['nb_hidden_units'])
elif config['model'] == 'mlp':
model = models.MLPModel(input_shape=(config['history_length'], ),
nb_output_units=1,
nb_hidden_units=config['nb_hidden_units'],
nb_layers=config['nb_layers'])
elif config['model'] == 'gru':
model = models.GRUModel(input_shape=(config['history_length'], 1),
nb_output_units=1,
nb_hidden_units=config['nb_hidden_units'],
nb_layers=config['nb_layers'],
dropout=config['dropout'],
recurrent_dropout=config['recurrent_dropout'])
elif config['model'] == 'lstm':
model = models.LSTMModel(input_shape=(config['history_length'], 1),
results += [{"Train score": trainscore, "Test score": testscore}]
results = pd.DataFrame(results)
results.plot()
plt.show()
elif part == '2':
data = du.load_dataset("digits")
X = data['X']
y = data['y']
Xtest = data['Xtest']
ytest = data['ytest']
model = models.LinearModel(n_features=X.shape[1], n_outputs=10)
results = []
for i in range(50):
model.fit(X,
y,
epochs=10,
batch_size=100,
verbose=0,
optimizer_name="adam",
learning_rate=1e-3,
weights_name="linaer")
# EVALUATE TRAIN AND TEST CLASSIFICATION
yhat = np.argmax(model.predict(X), axis=1)
trainscore = (yhat == y).mean()
