def test_learn_and_predict_nnmixer(self):
config = {
'input_features': [{
'name': 'sqft',
'type': 'numeric'
}, {
'name': 'days_on_market',
'type': 'numeric'
}, {
'name': 'neighborhood',
'type': 'categorical',
'dropout': 0.4
}],
'output_features': [{
'name': 'number_of_rooms',
'type': 'categorical',
'weights': {
'0': 0.8,
'1': 0.6,
'2': 0.5,
'3': 0.7,
'4': 1
}
}, {
'name': 'number_of_bathrooms',
'type': 'categorical',
'weights': {
'0': 0.8,
'1': 0.6,
'2': 4
}
}, {
'name': 'rental_price',
'type': 'numeric'
}, {
'name': 'location',
'type': 'categorical'
}],
'mixer': {
'class': NnMixer,
'kwargs': {
'eval_every_x_epochs': 4,
'stop_training_after_seconds': 10
}
}
}
df = pd.read_csv(
'https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv'
)
predictor = Predictor(config)
predictor.learn(from_data=df)
df = df.drop([x['name'] for x in config['output_features']], axis=1)
predictor.predict(when_data=df)
def test_multiple_categories_as_output(self):
pass # fails: AssertionError: 0.0 not greater than or equal to 0.15
vocab = self.get_vocab(10)
# x1 contains the index of first tag present
# x2 contains the index of second tag present
# if a tag is missing then x1/x2 contain -1 instead
# Thus the dataset should be perfectly predicted
n_points = 10000
x1 = [random.randint(0, len(vocab) - 1) if random.random() > 0.2 else -1 for i in range(n_points)]
x2 = [random.randint(0, len(vocab) - 1) if random.random() > 0.2 else -1 for i in range(n_points)]
tags = []
for x1_index, x2_index in zip(x1, x2):
row_tags = set([vocab.get(x1_index), vocab.get(x2_index)])
row_tags = [x for x in row_tags if x is not None]
tags.append(row_tags)
df = pd.DataFrame({'x1': x1, 'x2': x2, 'tags': tags})
config = {
'input_features': [
{'name': 'x1', 'type': ColumnDataTypes.CATEGORICAL},
{'name': 'x2', 'type': ColumnDataTypes.CATEGORICAL}
],
'output_features': [
{'name': 'tags', 'type': ColumnDataTypes.MULTIPLE_CATEGORICAL}
],
'mixer': {'class': NnMixer, 'kwargs': {'stop_training_after_seconds': 25}}
}
df_train = df.iloc[:round(n_points * 0.9)]
df_test = df.iloc[round(n_points * 0.9):]
predictor = Predictor(config)
predictor.learn(from_data=df_train)
predictions = predictor.predict(when_data=df_train)
train_tags = df_train.tags
predicted_tags = predictions['tags']['predictions']
train_tags_encoded = predictor._mixer.encoders['tags'].encode(train_tags)
pred_labels_encoded = predictor._mixer.encoders['tags'].encode(predicted_tags)
score = f1_score(train_tags_encoded, pred_labels_encoded, average='weighted')
print('Train f1 score', score)
self.assertGreaterEqual(score, 0.15)
predictions = predictor.predict(when_data=df_test)
test_tags = df_test.tags
predicted_tags = predictions['tags']['predictions']
test_tags_encoded = predictor._mixer.encoders['tags'].encode(test_tags)
pred_labels_encoded = predictor._mixer.encoders['tags'].encode(predicted_tags)
score = f1_score(test_tags_encoded, pred_labels_encoded, average='weighted')
print('Test f1 score', score)
self.assertGreaterEqual(score, 0.15)
def test_home_rentals(self):
lightwood.config.config.CONFIG.USE_CUDA = USE_CUDA
lightwood.config.config.CONFIG.PLINEAR = PLINEAR
config = {
'input_features': [
{'name': 'sqft', 'type': 'numeric'},
{'name': 'days_on_market', 'type': 'numeric'},
{'name': 'neighborhood', 'type': 'categorical', 'dropout': 0.4}
],
'output_features': [
{'name': 'number_of_rooms', 'type': 'categorical', 'weights': {'0': 0.8, '1': 0.6, '2': 0.5, '3': 0.7, '4': 1}},
{'name': 'number_of_bathrooms', 'type': 'categorical', 'weights': {'0': 0.8, '1': 0.6, '2': 4}},
{'name': 'rental_price', 'type': 'numeric'},
{'name': 'location', 'type': 'categorical'}
],
'data_source': {'cache_transformed_data': CACHE_ENCODED_DATA},
'mixer': {
'class': NnMixer,
'kwargs': {
'selfaware': SELFAWARE,
'eval_every_x_epochs': 4,
'stop_training_after_seconds': 80
}
}
}
df = pd.read_csv('https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv')
predictor = Predictor(config)
predictor.learn(from_data=df)
df = df.drop([x['name'] for x in config['output_features']], axis=1)
predictor.predict(when_data=df)
predictor.save('test.pkl')
predictor = Predictor(load_from_path='test.pkl')
for j in range(100):
pred = predictor.predict(when={'sqft': round(j * 10)})['number_of_rooms']['predictions'][0]
assert isinstance(pred, (str, int))
def test_multiple_categories_as_input(self):
vocab = self.get_vocab(10)
# tags contains up to 2 randomly selected tags
# y contains the sum of indices of tags
# the dataset should be nearly perfectly predicted
n_points = 10000
tags = []
y = []
for i in range(n_points):
row_tags = []
row_y = 0
for k in range(2):
if random.random() > 0.2:
selected_index = random.randint(0, len(vocab)-1)
if vocab[selected_index] not in row_tags:
row_tags.append(vocab[selected_index])
row_y += selected_index
tags.append(row_tags)
y.append(row_y)
df = pd.DataFrame({'tags': tags, 'y': y})
config = {
'input_features': [
{'name': 'tags', 'type': ColumnDataTypes.MULTIPLE_CATEGORICAL}
],
'output_features': [
{'name': 'y', 'type': ColumnDataTypes.NUMERIC}
],
'mixer': {
'class': NnMixer,
'kwargs': {'stop_training_after_seconds': 10}
}
}
df_train = df.iloc[:round(n_points * 0.9)]
df_test = df.iloc[round(n_points * 0.9):]
predictor = Predictor(config)
predictor.learn(from_data=df_train)
predictions = predictor.predict(when_data=df_test)
test_y = df_test.y
predicted_y = predictions['y']['predictions']
score = r2_score(test_y, predicted_y)
print('Test R2 score', score)
# The score check is very light because we only allow the model to train for a few seconds
# We are just checking that it learns something and predicts properly, not benchmarking here
self.assertGreaterEqual(score, 0.15)
'name': 'next',
'type': 'numeric'
}]
}
def iter_function(epoch, error, test_error, test_error_gradient):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, accuracy: {accuracy}'
.format(iter=epoch,
error=error,
test_error=test_error,
test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy))
data = pandas.DataFrame(ts_data, columns=['time', 'ts', 'next'])
predictor = Predictor(config)
predictor.learn(from_data=data,
callback_on_iter=iter_function,
eval_every_x_epochs=10)
ret = predictor.predict(when={
'ts':
" ".join([str(math.sin(i / max)) for i in range(10 + 1, 10 + ts_len)])
})
print(" ".join(
[str(math.sin(i / max)) for i in range(10 + 1, 10 + ts_len + 1)]))
print(ret)
import pandas as pd
from lightwood import Predictor
####################
config = {'input_features': [{'name': 'number_of_rooms', 'type': 'numeric'},
{'name': 'number_of_bathrooms', 'type': 'numeric'}, {'name': 'sqft', 'type': 'numeric'},
{'name': 'location', 'type': 'categorical'}, {'name': 'days_on_market', 'type': 'numeric'},
{'name': 'neighborhood', 'type': 'categorical'}],
'output_features': [{'name': 'rental_price', 'type': 'numeric'}]}
df=pd.read_csv("https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv")
predictor = Predictor(config)
def iter_function(epoch, error, test_error, test_error_gradient):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, accuracy: {accuracy}'.format(
iter=epoch, error=error, test_error=test_error, test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy))
predictor.learn(from_data=df, callback_on_iter=iter_function, eval_every_x_epochs=10)
print(predictor.predict(when={'number_of_rooms':3, 'number_of_bathrooms':2, 'sqft':700, 'location':'great'}))
def iter_function(epoch, error, test_error, test_error_gradient, test_accuracy):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, accuracy: {accuracy}, test_accuracy: {test_accuracy}'.format(
iter=epoch, error=error, test_error=test_error, test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy, test_accuracy=test_accuracy))
config = config_rp
# predictor = Predictor(config)
# predictor.learn(from_data=df, callback_on_iter=iter_function, eval_every_x_epochs=10)
# predictor.save('/tmp/predictor.mdb')
predictor = Predictor(load_from_path=file)
print(predictor.overall_certainty)
ret = []
for i in range(int(100-predictor.overall_certainty*100)):
ret_val = predictor.predict(when={'number_of_bedrooms': 2, 'sqft': 2300})[
config['output_features'][0]['name']]['predictions'][0]
if ret_val is not None:
ret += [ret_val]
hist, bin_edges = numpy.histogram(ret)
net = sum(hist)
for j, count in enumerate(hist):
print('{edge:.2f}: {value:.2%}'.format(value=count/net, edge=bin_edges[j]))
def run_full_test(USE_CUDA, CACHE_ENCODED_DATA, SELFAWARE, PLINEAR):
'''
Run full test example with home_rentals dataset
'''
lightwood.config.config.CONFIG.USE_CUDA = USE_CUDA
lightwood.config.config.CONFIG.PLINEAR = PLINEAR
config = {
'input_features': [{
'name': 'number_of_bathrooms',
'type': 'numeric'
}, {
'name': 'sqft',
'type': 'numeric'
}, {
'name': 'location',
'type': 'categorical'
}, {
'name': 'days_on_market',
'type': 'numeric'
}, {
'name': 'neighborhood',
'type': 'categorical',
'dropout': 0.4
}, {
'name': 'rental_price',
'type': 'numeric'
}],
'output_features': [{
'name': 'number_of_rooms',
'type': 'categorical',
# 'weights':{
# '0': 0.8,
# '1': 0.6,
# '2': 0.5,
# '3': 0.7,
# '4': 1,
# }
}],
'data_source': {
'cache_transformed_data': CACHE_ENCODED_DATA
},
'mixer': {
'class': lightwood.BUILTIN_MIXERS.NnMixer,
'selfaware': SELFAWARE
}
}
df = pd.read_csv(
"https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv"
)
def iter_function(epoch, error, test_error, test_error_gradient,
test_accuracy):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, test_accuracy: {test_accuracy}'
.format(iter=epoch,
error=error,
test_error=test_error,
test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy,
test_accuracy=test_accuracy))
predictor = Predictor(config)
# stop_training_after_seconds given in order to not get timeouts in travis
predictor.learn(from_data=df,
callback_on_iter=iter_function,
eval_every_x_epochs=4,
stop_training_after_seconds=40)
df = df.drop([x['name'] for x in config['output_features']], axis=1)
predictor.predict(when_data=df)
predictor.save('test.pkl')
predictor = Predictor(load_from_path='test.pkl')
preds = {}
for j in range(100):
pred = predictor.predict(
when={'sqft': round(j * 10)})['number_of_rooms']['predictions'][0]
if pred not in preds:
preds[pred] = 0
preds[pred] += 1
}],
'output_features': [{
'name': 'superclass',
'type': 'categorical',
'encoder_attrs': {}
}, {
'name': 'class',
'type': 'categorical',
'encoder_attrs': {}
}]
}
predictor = Predictor(config)
def iter_function(epoch, error, test_error, test_error_gradient):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, accuracy: {accuracy}'
.format(iter=epoch,
error=error,
test_error=test_error,
test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy))
predictor.learn(from_data=pd.read_csv('train_sample.csv'),
callback_on_iter=iter_function,
eval_every_x_epochs=1)
results = predictor.predict(when_data=pd.read_csv('test_sample.csv'))
print(results)
'mixer': {
'class': lightwood.BUILTIN_MIXERS.BayesianNnMixer
}
}
#'mixer':{'class': lightwood.BUILTIN_MIXERS.NnMixer}}
df = pd.read_csv(
"https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv")
predictor = Predictor(config)
def iter_function(epoch, error, test_error, test_error_gradient,
test_accuracy):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, test_accuracy: {test_accuracy}'
.format(iter=epoch,
error=error,
test_error=test_error,
test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy,
test_accuracy=test_accuracy))
predictor.learn(from_data=df,
callback_on_iter=iter_function,
eval_every_x_epochs=2,
stop_training_after_seconds=20)
print(predictor.predict(when_data=df.iloc[0:20]))
#print(predictor.predict(when={'number_of_rooms':3, 'number_of_bathrooms':2, 'sqft':700, 'location':'great'}))
def iter_function(epoch, error, test_error, test_error_gradient, test_accuracy):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, test_accuracy: {test_accuracy}'.format(
iter=epoch, error=error, test_error=test_error, test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy, test_accuracy=test_accuracy))
predictor.learn(from_data=data_frame, callback_on_iter=iter_function)
print('accuracy')
print(predictor.train_accuracy)
print('accuracy over all dataset')
print(predictor.calculate_accuracy(from_data=data_frame))
when = {'x': [1], 'y': [0]}
print('- multiply when. {when}'.format(when=when))
print(predictor.predict(when=when))
# saving the predictor
predictor.save('ok.pkl')
# loading the predictor
predictor2 = Predictor(load_from_path='ok.pkl')
when = {'x': [0, 0, 1, -1, 1], 'y': [0, 1, -1, -1, 1]}
print('- multiply when. {when}'.format(when=when))
print(predictor2.predict(when_data=pandas.DataFrame(when)))
when = {'x': [0, 3, 1, -5, 1], 'y': [0, 1, -5, -4, 7]}
print('- multiply when. {when}'.format(when=when))
print(predictor2.predict(when_data=pandas.DataFrame(when)))
config = {
'input_features': [{
'name': 'ts',
'type': COLUMN_DATA_TYPES.TIME_SERIES
}],
'output_features': [{
'name': 'next',
'type': 'numeric'
}]
}
def iter_function(epoch, training_error, test_error, test_error_gradient,
test_accuracy):
print(
f'Epoch: {epoch}, Train Error: {training_error}, Test Error: {test_error}, Test Error Gradient: {test_error_gradient}, Test Accuracy: {test_accuracy}'
)
data = pandas.DataFrame(ts_data, columns=['time', 'ts', 'next'])
predictor = Predictor(config)
predictor.learn(from_data=data)
print('\n\n')
ret = predictor.predict(
when={'ts': [math.sin(i / max) for i in range(10 + 1, 10 + ts_len)]})
print([math.sin(i / max) for i in range(10 + 1, 10 + ts_len + 1)])
print('Got predictions: ')
print(ret)
def iter_function(epoch, training_error, test_error, test_error_gradient,
test_accuracy):
print(
f'Epoch: {epoch}, Train Error: {training_error}, Test Error: {test_error}, Test Error Gradient: {test_error_gradient}, Test Accuracy: {test_accuracy}'
)
predictor.learn(from_data=df_train,
callback_on_iter=iter_function,
eval_every_x_epochs=200)
predictor.save('ok.pkl')
predictor = Predictor(load_from_path='ok.pkl')
print('Train accuracy: ', predictor.train_accuracy)
print('Test accuracy: ', predictor.calculate_accuracy(from_data=df_test))
predictions = predictor.predict(when_data=df_test)
print('Confidence mean for both x and y present: ',
np.mean(predictions['z']['selfaware_confidences']))
print(list(df_test['z'])[30:60])
print(predictions['z']['predictions'][30:60])
predictions = predictor.predict(when_data=df_test.drop(columns=['x']))
print('Confidence mean for x missing: ',
np.mean(predictions['z']['selfaware_confidences']))
predictions = predictor.predict(when_data=df_test.drop(columns=['y']))
print('Confidence mean for y missing: ',
np.mean(predictions['z']['selfaware_confidences']))
def run_test(USE_CUDA, CACHE_ENCODED_DATA, SELFAWARE, PLINEAR):
lightwood.config.config.CONFIG.USE_CUDA = USE_CUDA
lightwood.config.config.CONFIG.CACHE_ENCODED_DATA = CACHE_ENCODED_DATA
lightwood.config.config.CONFIG.SELFAWARE = SELFAWARE
lightwood.config.config.CONFIG.PLINEAR = PLINEAR
####################
config = {
'input_features': [{
'name': 'number_of_bathrooms',
'type': 'numeric'
}, {
'name': 'sqft',
'type': 'numeric'
}, {
'name': 'location',
'type': 'categorical'
}, {
'name': 'days_on_market',
'type': 'numeric'
}, {
'name': 'neighborhood',
'type': 'categorical',
'dropout': 0.4
}, {
'name': 'rental_price',
'type': 'numeric'
}],
'output_features': [{
'name': 'number_of_rooms',
'type': 'categorical',
# 'weights':{
# '0': 0.8,
# '1': 0.6,
# '2': 0.5,
# '3': 0.7,
# '4': 1,
# }
}],
'mixer': {
'class': lightwood.BUILTIN_MIXERS.NnMixer
}
}
# AX doesn't seem to work on the travis version of windows, so don't test it there as of now
if sys.platform not in ['win32', 'cygwin', 'windows']:
pass
#config['optimizer'] = lightwood.model_building.BasicAxOptimizer
df = pd.read_csv(
"https://mindsdb-example-data.s3.eu-west-2.amazonaws.com/home_rentals.csv"
)
def iter_function(epoch, error, test_error, test_error_gradient,
test_accuracy):
print(
'epoch: {iter}, error: {error}, test_error: {test_error}, test_error_gradient: {test_error_gradient}, test_accuracy: {test_accuracy}'
.format(iter=epoch,
error=error,
test_error=test_error,
test_error_gradient=test_error_gradient,
accuracy=predictor.train_accuracy,
test_accuracy=test_accuracy))
predictor = Predictor(config)
# stop_training_after_seconds given in order to not get timeouts in travis
predictor.learn(from_data=df,
callback_on_iter=iter_function,
eval_every_x_epochs=1,
stop_training_after_seconds=1)
predictor.save('test.pkl')
predictor = Predictor(load_from_path='test.pkl')
df = df.drop([x['name'] for x in config['output_features']], axis=1)
predictor.predict(when_data=df)
predictor.save('test.pkl')
predictor = Predictor(load_from_path='test.pkl')
preds = {}
for j in range(100):
pred = predictor.predict(
when={'sqft': round(j * 10)})['number_of_rooms']['predictions'][0]
if pred not in preds:
preds[pred] = 0
preds[pred] += 1