QuickBI
I. Introduction
1. Overview: huge data-line analysis, the drag operation, commanded visualization. Data analysis can be done by a simple operation, traffic data profiling.
2. Features:
(1) to support multiple types of data sources, supports multiple visual components
real-time analysis (2) massive amounts of data, providing one-click intelligent acceleration
(3) flexible reporting integrated solutions, tight security Rights Management
(4 ) low threshold approachable save time, lower cost-saving cloud computing costs
3. architecture
4. solution role in big data
5. common application step
two, data management
1. data source Administrator:
(1) cloud data sources: MaxCompute, AnaliyticDB, RDS (Mysql, SQLServer), Hybrid
DB (Greenplum)
(2) self-built data sources: MySQL, sqlserver
(3) exploration of space: a local csv file, local excel file
2. data set management: toolbar data panel, dimensions panel, metrics panel
(1) operations for the dimension that can be performed: edit, delete, Kelong Wei degree, create a management hierarchy, converted to metric, the new calculated field (dimension), dimension type switch (by default, the date, geographic information )
(2) for the operation of the metric may be: edit, delete, create a calculated field, one moving, into dimensional, digital format, default aggregation ( And (default), count / duplication technology, the maximum / minimum / average)
three, icons portal
1. icon
2. Report portal
(1) Portal: data products, are set through a menu in the form of an instrument panel of tissue can be produced through a complicated menu with navigation data portals thematic class analysis
Master machine learning PAI
First, the machine learning Introduction
1. Definitions:
(1) without programming directly to the question, giving a field of computer research learning ability
(2) of a machine, a large number of historical data by statistical learning algorithms to generate empirical model, using the experience of our business model
2. checkers and machine learning:
(1) brute force:
[1] methods:
a should form the best hand in all situations, as the knowledge base.
b traverse from the knowledge base current. situation, according to the best hand row should move
[2] problems:
. A knowledge base is difficult to produce, a huge amount of data
b iterate through the data from the mass.
(2) fixed routine
[1] method:
. A summary of some of the rules of chess line
. b according to line chess rules line chess
[2] problem:
. a line chess rules summarize trouble
b human level playing strength and summarize the rules concerned.
(3) training model:
[1] thinking machine learning of
a task T:. voyages Checkers
b performance criteria P:. win at the probability
c experience E:. and their chess
d objective function:. V
[2] evaluation chess game status b:
a.x1: the number of sunspots chess
b.x2: the chess the number of sub-red
number on the board of the black king: c.x3
d.x4: red Net on the board Number
e.x5: the number of sunspots is red sub-threatening
f.x6: the number of sunspots red threat
[3] ideas:
B1-> B2-> ... -> End
input or game record discs themselves and their many, can be obtained w to, W1 ... W6 of
V (B) X1 + W0 = W1 + X6 + ... W6 of
(. 4) summarizes
[1] features:
. a without the traditional programming mode
. b defined task, performance, experience, and the objective function, the objective function and learning can be provided
. c with data changes, automatically learning, updating
d using their chess => manner model optimization, can continue to improve.
[2] Application scenario:
. a not manually programmed for issue
b can not define a solution to this problem.
C based on complex data. rapid decision-making
. d massive personalization system
2. common classification - learning in different ways according to the classification: Depending on the study data, the problems there are different modeling methods
(1) supervised learning: learning the outcome of the sample mark - classification return
(2) unsupervised learning: learning samples inconclusive mark
[1] from no training data labeled a conclusion. The characteristics of the input data does not exist after the clear identification of the results.
[2] as a common cluster: find hidden patterns or data packets.
(3) semi-supervised learning: Learning samples with a portion of the recording mark results
two, of PAI Introduction
1. Introduction: Large machine learning based internet MaxCompute provide data processing, modeling, prediction offline, the online prediction and services.
2. Features:
(1) easy to use: the package bottom Tonggu distributed algorithm for providing visual drag pull operating environment,
rich in (2) Algorithm: providing engineering characteristics, pre-processing data, statistical analysis, machine learning, learning framework depth, prediction estimation and the like algorithmic components in more than 100
(3) one-stop experience: provides online forecasts and Lee now scheduling function, allows the machine training results and business can seamlessly
(4) deep learning: currently supports MXNET, CAFFE, TENSORFLOW three mainstream depth learning framework, the underlying platform to provide training gpu
3. advantages
(1) storage and computational costs of: computing and storing are used
(2) a large number of tools, algorithm library, reduced technical barriers
(3) visualization of the drag operation, complete solution
4. product composition
5. algorithm supported:
