def load_dataset(filename):
"""
Load given file with SFrame
"""
data_frame = sframe.SFrame()
train_data = sframe.SFrame()
dataset = data_frame.read_csv(filename,
delimiter='|',
header=False,
nrows=100)
train_data['topics'] = dataset['X1']
train_data['content'] = dataset['X2']
return train_data, dataset['X3']
def query(vec, model, k, max_search_radius):
data = model['data']
table = model['table']
random_vectors = model['random_vectors']
num_vector = random_vectors.shape[1]
# Compute bin index for the query vector, in bit representation.
bin_index_bits = (vec.dot(random_vectors) >= 0).flatten()
# Search nearby bins and collect candidates
candidate_set = set()
for search_radius in xrange(max_search_radius + 1):
candidate_set = search_nearby_bins(bin_index_bits,
table,
search_radius,
initial_candidates=candidate_set)
# Sort candidates by their true distances from the query
nearest_neighbors = sframe.SFrame({'id': candidate_set})
candidates = data[np.array(list(candidate_set)), :]
nearest_neighbors['distance'] = pairwise_distances(
candidates, vec, metric='cosine').flatten()
return nearest_neighbors.topk('distance', k,
reverse=True), len(candidate_set)
def test_zero_size_array(self):
self.data = gl.SFrame()
self.data['x'] = [array.array('d')] * 10
it = mxnet.io.SFrameIter(self.data, data_field='x')
data_actual = []
for d in it:
data_actual.extend(d.data[0].asnumpy().flatten())
self.assertEquals(data_actual, [])
def test_variable_size_image(self):
shape1 = (2, 3, 1)
shape2 = (2, 2, 2)
tmp1 = gl.SArray([array.array('d', [0] * 6)])
tmp2 = gl.SArray([array.array('d', [0] * 8)])
data = gl.SFrame({'x': [tmp1.pixel_array_to_image(*shape1)[0], tmp2.pixel_array_to_image(*shape2)[0]]})
it = mxnet.io.SFrameIter(data, data_field='x')
self.assertRaises(lambda: [it])
def brute_force_query(vec, data, k):
num_data_points = data.shape[0]
# Compute distances for ALL data points in training set
nearest_neighbors = sframe.SFrame({'id':range(num_data_points)})
nearest_neighbors['distance'] = metrics.pairwise_distances(data, vec, metric='cosine').flatten()
return nearest_neighbors.topk('distance', k, reverse=True)
def setUp(self):
self.data = gl.SFrame({'x': [np.random.randn(8)] * 10,
'y': np.random.randint(2, size=10)})
self.shape = (8,)
self.label_field = 'y'
self.data_field = 'x'
self.data_size = len(self.data)
self.data_expected = list(x for arr in self.data['x'] for x in arr)
self.label_expected = list(self.data['y'])
def setUp(self):
self.data = gl.SFrame({'x': np.random.randn(10),
'y': np.random.randint(2, size=10)})
self.shape = (1,)
self.label_field = 'y'
self.data_field = 'x'
self.data_size = len(self.data)
self.data_expected = list(self.data['x'])
self.label_expected = list(self.data['y'])
def setUp(self):
(w, h, c, n) = (2, 4, 3, 100)
self.images = [np.random.randint(256, size=(h,w,c)) for i in range(n)]
self.data = gl.SFrame({'arr': [array.array('d', x.flatten()) for x in self.images],
'y': np.random.randint(2, size=n)})
self.data['img'] = self.data['arr'].pixel_array_to_image(w, h, c)
self.shape = (c, h, w)
self.label_field = 'y'
self.data_field = 'img'
self.data_size = len(self.data)
self.data_expected = list(x for arr in self.data['arr'] for x in arr)
self.label_expected = list(self.data['y'])
def polynomial_sframe(feature, degree):
# assume that degree >= 1 & initialize the SFrame
poly_sframe = sframe.SFrame()
# first degree
poly_sframe['power_1'] = feature
if degree > 1:
# then loop over the remaining degrees:
# range usually starts at 0 and stops at the endpoint-1. We want it to start at 2 and stop at degree
for power in range(2, degree + 1):
name = 'power_' + str(power)
poly_sframe[name] = feature.apply(lambda x: x**power)
return poly_sframe
def test_size1_array(self):
# setup data
self.data = gl.SFrame({'x': [np.random.randn(1)] * 10,
'y': np.random.randint(2, size=10)})
self.shape = (1,)
self.label_field = 'y'
self.data_field = 'x'
self.data_size = len(self.data)
self.data_expected = list(x for arr in self.data['x'] for x in arr)
self.label_expected = list(self.data['y'])
self.test_one_batch()
self.test_non_divisible_batch()
self.test_padding()
self.test_shape_inference()
def CreateDataFrame(csvFileName, protocol, sframe):
if sframe:
import sframe
frameName = csvFileName.replace(".csv", "_SFRAME")
dataframe = sframe.SFrame(csvFileName) #create dataframe in SFrame
dataframe.save(frameName) #save sframe
print dat(), "Creating SFRAME:", frameName
else:
import pandas
frameName = csvFileName.replace(".csv", ".PANDAS")
pDataframe = pandas.read_csv(csvFileName).fillna(
'N/A') #create pandas dataframe
pDataframe.to_pickle(frameName) #save pandas dataframe
print dat(), "Creating:", frameName
def read_and_clean_data():
'''Reads data and clean the words with punctuation
e.g. Hello! => Hello
'''
products = sframe.SFrame('amazon_baby.gl/')
products['clean_review'] = products['review'].apply(remove_punctuation)
# Filter the neutral ratings, neutral ratings does not help much in
# learning process
products = products[products['rating'] != 3]
# Add sentiment feature to indicate whether reviwe is positive or negative
# this will be out target of prediction
products['sentiment'] = products['rating'].apply(lambda rating : \
1 if rating > 3 else -1)
return products
def get_numpy_data(data_sframe, features, output):
# add a constant column to an SFrame
data_sframe['constant'] = 1
# prepend variable 'constant' to the features list
features = ['constant'] + features
# select the columns of data_SFrame given by the 'features' list into the SFrame 'features_sframe'
features_sframe = sframe.SFrame()
features_sframe[features] = data_sframe[features]
# this will convert the features_sframe into a numpy matrix with GraphLab Create >= 1.7!!
features_matrix = features_sframe.to_numpy()
features_matrix = features_matrix
# assign the column of data_sframe associated with the target to the variable 'output_sarray'
output_sarray = data_sframe[output]
# this will convert the SArray into a numpy array:
output_array = output_sarray.to_numpy() # GraphLab Create>= 1.7!!
return features_matrix, output_array
def setUp(self):
w = 2
h = 3
c = 1
d = 6
n = 5
self.data = gl.SFrame({'arr': [array.array('d', range(0, 6)),
array.array('d', range(50, 56)),
array.array('d', range(100, 106)),
array.array('d', range(200, 206)),
array.array('d', range(249, 255))],
'y': np.random.randint(2, size=n)})
self.data['img'] = self.data['arr'].pixel_array_to_image(w, h, c)
self.shape = (c, h, w)
self.label_field = 'y'
self.data_field = 'img'
self.data_size = len(self.data)
self.data_expected = list(x for arr in self.data['arr'] for x in arr)
self.label_expected = list(self.data['y'])
def setUp(self):
self.data = gl.SFrame({'i': [x for x in range(10)],
'-i': [-x for x in range(10)],
'f': [float(x) for x in range(10)],
'-f': [-float(x) for x in range(10)],
'arr': [range(2) for x in range(10)],
'y': np.random.randint(2, size=10)})
self.shape = (6,)
self.label_field = 'y'
self.data_field = ['i', '-i', 'f', '-f', 'arr']
self.data_size = len(self.data)
def val_iter():
for row in self.data:
for col in self.data_field:
v = row[col]
if type(v) is array.array:
for x in v:
yield x
else:
yield float(v)
self.data_expected = list(val_iter())
self.label_expected = list(self.data['y'])
def lag(df, features, numlag, type='pd'):
if len(features) == len(numlag):
if type == 'sf':
import sframe as SF
sf = SF.SFrame(df)
for j in range(0, len(features)):
feature = features[j]
column = sf[feature]
for i in range(1, numlag[j] + 1):
lead = [0] * i
stri = str(i)
lead.extend(column[0:len(column) - i])
exec('sf["' + feature + 'lag' + stri + '"] = lead')
sf = sf[max(numlag):len(sf)]
return (sf)
elif type == 'pd':
import pandas as pd
sf = pd.DataFrame(df)
for j in range(0, len(features)):
feature = features[j]
column = sf[feature]
for i in range(1, numlag[j] + 1):
lead = [0] * i
stri = str(i)
lead.extend(column[0:len(column) - i])
exec('sf["' + feature + 'lag' + stri + '"] = lead')
sf = sf[max(numlag):len(sf)]
return (sf)
else:
print('len(features) != len(numlag)')
return (0)
import sframe
loans = sframe.SFrame('lending-club-data.gl/')
features = ['grade', # grade of the loan
'term', # the term of the loan
'home_ownership', # home_ownership status: own, mortgage or rent
'emp_length', # number of years of employment
]
target = 'safe_loans'
loans['safe_loans'] = loans['bad_loans'].apply(lambda x : +1 if x==0 else -1)
loans = loans.remove_column('bad_loans')
# Subsample dataset to make sure classes are balanced
safe_loans_raw = loans[loans[target] == 1]
risky_loans_raw = loans[loans[target] == -1]
# Since there are less risky loans than safe loans, find the ratio of the sizes
# and use that percentage to undersample the safe loans.
percentage = len(risky_loans_raw)/float(len(safe_loans_raw))
safe_loans = safe_loans_raw.sample(percentage, seed = 1)
risky_loans = risky_loans_raw
loans_data = risky_loans.append(safe_loans)
print "Percentage of safe loans :", len(safe_loans) / float(len(loans_data))
print "Percentage of risky loans :", len(risky_loans) / float(len(loans_data))
print "Total number of loans in our new dataset :", len(loans_data)
loans_data_features = loans_data[features]
def one_hot_normalize_to_columns(sfData):
word_count_table = row[['tf_idf'
]].stack('tf_idf',
new_column_name=['word', 'weight'])
return word_count_table.sort('weight', ascending=False)
def has_top_words(word_count_vector):
# extract the keys of word_count_vector and convert it to a set
unique_words = set(word_count_vector.keys())
# return True if common_words is a subset of unique_words
# return False otherwise
return common_words.issubset(unique_words)
# El cuerpo de todos los datos de wikipedia sobre los que trabajaremos
wiki = sframe.SFrame('..//w2-a1//people_wiki.gl/')
wiki = wiki.add_row_number()
# El conteo de palabras de cada articulo nos lo dan, aunque se podria extraer por nuestros propios medios (explorar sklearn.CountVectorize)
word_count = load_sparse_csr('..//w2-a1//people_wiki_word_count.npz')
map_index_to_word = sframe.SFrame(
'..//w2-a1//people_wiki_map_index_to_word.gl/')
# Ahora experimentaremos con KNN. Primero con los word counts en Bruto
model = NearestNeighbors(metric='euclidean', algorithm='brute')
model.fit(word_count)
# Buscamos el artculo mas parecido a Obama
obama_id = wiki[wiki['name'] == 'Barack Obama']['id'][0]
distances, indices = model.kneighbors(word_count[obama_id], n_neighbors=10)
Created on Sun Dec 17 11:37:56 2017
@author: Abhishek S
"""
import sframe as sf
import numpy as np
import sframe.aggregate as agg
import sklearn
import sklearn.ensemble
from sklearn.ensemble import GradientBoostingClassifier
loans=sf.SFrame('E:/Machine learning Classification/Week 5/lending-club-data.gl')
loans
loans.print_rows(5,68)
dt={0:1,1:-1}
loans['safe_loans']=loans['bad_loans'].apply(lambda x:1 if x==0 else -1)
del loans['bad_loans']
loans
target = 'safe_loans'
features = ['grade', # grade of the loan (categorical)
'sub_grade_num', # sub-grade of the loan as a number from 0 to 1
'short_emp', # one year or less of employment
'emp_length_num', # number of years of employment
'home_ownership', # home_ownership status: own, mortgage or rent
'dti', # debt to income ratio
'purpose', # the purpose of the loan
'payment_inc_ratio', # ratio of the monthly payment to income
for i in xrange(num_doc)]
def top_words(name):
"""
Get a table of the most frequent words in the given person's wikipedia page.
"""
row = wiki[wiki['name'] == name]
print row
word_count_table = row[['word_count'
]].stack('word_count',
new_column_name=['word', 'count'])
return word_count_table.sort('count', ascending=False)
wiki = sframe.SFrame('people_wiki.gl/')
wiki = wiki.add_row_number()
word_count = load_sparse_csr('people_wiki_word_count.npz')
# print word_count[35817]
map_index_to_word = sframe.SFrame('people_wiki_map_index_to_word.gl/')
# model = NearestNeighbors(metric='euclidean', algorithm='brute')
# model.fit(word_count)
# print wiki[wiki['name'] == 'Barack Obama']
# distances, indices = model.kneighbors(word_count[35817], n_neighbors=10) # 1st
# print distances, indices
# neighbors = sframe.SFrame({'distance':distances.flatten(), 'id':indices.flatten()})
# print wiki.join(neighbors, on='id').sort('distance')[['id','name','distance']]
wiki['word_count'] = unpack_dict(word_count, map_index_to_word)
# neogi - just numpy functions BEST ONE
# ramaranjanruj - uses sklearn
# corylstewart - uses graphlab
# justindomingue - very high level only functions
import pandas as pd
import sframe
import numpy as np
from sklearn.model_selection import train_test_split
from math import sqrt
#import graphlab
# import data
sales = sframe.SFrame(
'/Users/davidbartram-shaw/Machine Learning Course/Course 2 - Regression/kc_house_data.gl/'
)
#sales=sframe.SFrame.to_dataframe(sales)
# Test/Train split
(train_and_validation,
test) = sales.random_split(.8, seed=1) # initial train/test split
(train, validation) = train_and_validation.random_split(
.8, seed=1) # split training set into training and validation sets
#################################################################
# REGRESSION - KNN REGRESSION & KERNALS
#################################################################
# optimize matrix operations
# brute force query, for comparison
def brute_force_query(vec, data, k):
num_data_points = data.shape[0]
# Compute distances for ALL data points in training set
nearest_neighbors = sframe.SFrame({'id':range(num_data_points)})
nearest_neighbors['distance'] = metrics.pairwise_distances(data, vec, metric='cosine').flatten()
return nearest_neighbors.topk('distance', k, reverse=True)
# -------------------------------------------------------------------------------
# Test #
# -------------------------------------------------------------------------------
people = sframe.SFrame('data/people_wiki.gl')
people = people.add_row_number()
# preprocessed tf-idf vectors
words = load_sparse_csr('data/people_wiki_tf_idf.npz')
map_index_to_word = sframe.SFrame('data/people_wiki_map_index_to_word.gl')
model = train_lsh(words, num_vector=16, seed=143)
example_search = 'Alberto Contador'
example_search_document = people[people['name'] == example_search]
example_search_id = example_search_document['id'][0]
example_bin = [key for key, value in model['table'].iteritems() if example_search_id in value][0]
print('\nBin for example search({}): {}, content: {}\n'.format(example_search, example_bin, model['table'][example_bin]))
from sklearn.linear_model import LogisticRegression
import numpy as np
# Function to remove punctuations
def remove_punctuation(text):
return string.translate(text, None, string.punctuation)
# Function to compute sigmoid response
def sigmoid(x):
return 1.0 / (1 + np.exp(-x))
# Read data
products = sframe.SFrame('../data/Week01/amazon_baby.gl/')
products['review_clean'] = products['review'].apply(remove_punctuation)
# Discard rating with value 3; these are treated as neither negative nor positive
products = products[products['rating'] != 3]
products['sentiment'] = products['rating'].apply(lambda rating: +1
if rating > 3 else -1)
# Split data into training and testing sets
train_data, test_data = products.random_split(fraction=0.8, seed=1)
# Create training and test matrices from the corresponding data using a CountVectorizer
vectorizer = CountVectorizer(token_pattern=r'\b\w+\b')
train_matrix = vectorizer.fit_transform(train_data['review_clean'])
test_matrix = vectorizer.transform(test_data['review_clean'])
words = vectorizer.get_feature_names()
#!/usr/bin/env python
# coding: utf-8
# In[10]:
import sframe
# # Read product review data
# In[23]:
products = sframe.SFrame('../Data/amazon_baby.gl')
# In[33]:
pwd
# In[34]:
products.save('../Data/products.csv', format='csv')
# In[2]:
import sframe
import pandas as pd
import numpy as np
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
products = sframe.SFrame('amazon_baby.gl/')
def remove_punctuation(text):
import string
return text.translate(None, string.punctuation)
# method for printing out confusion matrix
# In the case of binary classification, the confusion matrix is a 2-by-2 matrix
def print_confusion_matrix(y, y_hat, classifier):
from sklearn.metrics import confusion_matrix
# use the same order of class as the LR model
cmat = confusion_matrix(y_true = y, y_pred = y_hat, labels = classifier.classes_)
print ' target_label | predicted_label | count '
print '--------------+-----------------+-------'
# Print out the confusion matrix.
for i, target_label in enumerate(classifier.classes_):
for j, predicted_label in enumerate(classifier.classes_):
print '{0:^13} | {1:^15} | {2:5d}'.format(target_label, predicted_label, cmat[i,j])
def apply_threshold(probabilities, threshold):
return probabilities.applymap(lambda x : 1 if x >= threshold else -1) ## !!
import sframe
import json
import numpy as np
from math import sqrt
products = sframe.SFrame('amazon_baby_subset.gl/')
# For this assignment, we eliminated class imbalance by choosing a
# subset of the data with a similar number of positive and negative reviews.
print '# of positive reviews =', len(products[products['sentiment'] == 1])
print '# of negative reviews =', len(products[products['sentiment'] == -1])
def remove_punctuation(text):
import string
return text.translate(None, string.punctuation)
# the following function extracts columns from an SFrame and converts
# them into a NumPy array
#
# the feature matrix includes an additional column 'intercept'
# to take account of the intercept term - all 1s
def get_numpy_data(data_sframe, features, label):
data_sframe['intercept'] = 1
features = ['intercept'] + features
features_sframe = data_sframe[features]
feature_matrix = features_sframe.to_numpy()
label_sarray = data_sframe[label]
label_array = label_sarray.to_numpy()
return (feature_matrix, label_array)
def test_variable_size_array(self):
self.data = gl.SFrame({'x': [[0], [0, 1], [0, 1, 2]]})
self.assertRaises(ValueError, lambda: mxnet.io.SFrameIter(self.data, data_field='x'))
# Function to compute accuracy
def accuracy(prediction, actual):
"""
Purpose: Compute accuracy
Input : Predicted output values, true output values
Output : Accuracy
"""
prediction_correct = sum((actual == prediction) * 1.0)
prediction_total = len(prediction)
accuracy = prediction_correct / prediction_total
return accuracy
# Read data
loans = sframe.SFrame('../data/Week05/lending-club-data.gl/')
# Preprocess data
loans['safe_loans'] = loans['bad_loans'].apply(lambda x: +1 if x == 0 else -1)
loans = loans.remove_column('bad_loans')
# Selected features
features = [
'grade', # grade of the loan (categorical)
'sub_grade_num', # sub-grade of the loan as a number from 0 to 1
'short_emp', # one year or less of employment
'emp_length_num', # number of years of employment
'home_ownership', # home_ownership status: own, mortgage or rent
'dti', # debt to income ratio
'purpose', # the purpose of the loan
'payment_inc_ratio', # ratio of the monthly payment to income
# Function to plot likelihood curves
def make_plot(log_likelihood_all, len_data, batch_size, smoothing_window=1, label=''):
plt.rcParams.update({'figure.figsize': (9,5)})
log_likelihood_all_ma = np.convolve(np.array(log_likelihood_all), \
np.ones((smoothing_window,))/smoothing_window, mode='valid')
plt.plot(np.array(range(smoothing_window-1, len(log_likelihood_all)))*float(batch_size)/len_data,
log_likelihood_all_ma, linewidth=4.0, label=label)
plt.rcParams.update({'font.size': 16})
plt.tight_layout()
plt.xlabel('# of passes over data')
plt.ylabel('Average log likelihood per data point')
plt.legend(loc='lower right', prop={'size':14})
# Read data
products = sframe.SFrame('../data/Week02/amazon_baby_subset.gl/')
# Set of important words; these will be the features
list_of_words = ["baby", "one", "great", "love", "use", "would", "like", "easy", "little", "seat", "old", "well", "get", "also", "really", "son", "time", "bought", "product", "good", "daughter", "much", "loves", "stroller", "put", "months", "car", "still", "back", "used", "recommend", "first", "even", "perfect", "nice", "bag", "two", "using", "got", "fit", "around", "diaper", "enough", "month", "price", "go", "could", "soft", "since", "buy", "room", "works", "made", "child", "keep", "size", "small", "need", "year", "big", "make", "take", "easily", "think", "crib", "clean", "way", "quality", "thing", "better", "without", "set", "new", "every", "cute", "best", "bottles", "work", "purchased", "right", "lot", "side", "happy", "comfortable", "toy", "able", "kids", "bit", "night", "long", "fits", "see", "us", "another", "play", "day", "money", "monitor", "tried", "thought", "never", "item", "hard", "plastic", "however", "disappointed", "reviews", "something", "going", "pump", "bottle", "cup", "waste", "return", "amazon", "different", "top", "want", "problem", "know", "water", "try", "received", "sure", "times", "chair", "find", "hold", "gate", "open", "bottom", "away", "actually", "cheap", "worked", "getting", "ordered", "came", "milk", "bad", "part", "worth", "found", "cover", "many", "design", "looking", "weeks", "say", "wanted", "look", "place", "purchase", "looks", "second", "piece", "box", "pretty", "trying", "difficult", "together", "though", "give", "started", "anything", "last", "company", "come", "returned", "maybe", "took", "broke", "makes", "stay", "instead", "idea", "head", "said", "less", "went", "working", "high", "unit", "seems", "picture", "completely", "wish", "buying", "babies", "won", "tub", "almost", "either"]
# The label
label = ['sentiment']
# Remove punctuations
products['review_clean'] = products['review'].apply(lr_mle_sg.remove_punctuation)
# For each important word add a new column and determine count of that word in all reviews
for word in list_of_words:
products[word] = products['review_clean'].apply(lambda x: x.split().count(word))
train_data, validation_data = products.random_split(.9, seed=1)
'sentiment': 1
}, {
'text': 'I hate this f*****g product. Piece of shit.',
'sentiment': -1
}, {
'text': 'I love this excellent product. Great great great',
'sentiment': 1
}, {
'text': 'Hate hate hate! Never again. Bad',
'sentiment': 1
}, {
'text': 'Bad product. Really bad. I hate it.',
'sentiment': -1
}]
sf = sframe.SFrame(train_dataset)
data = sf.unpack('X1')
count_vect = CountVectorizer()
counts = count_vect.fit_transform(data['X1.text'])
regression = LogisticRegression()
model = regression.fit(counts, data['X1.sentiment'])
predict_data = ["I love this product.", "This is a really bad product."]
predict_count_vects = CountVectorizer(vocabulary=count_vect.vocabulary_)
predict_counts = predict_count_vects.fit_transform(predict_data)
predictions = model.predict(predict_counts)
for text, prediction in zip(predict_data, predictions):
