I've written a simple recurrent network in TensorFlow based on this video that I watched: https://youtu.be/vq2nnJ4g6N0?t=8546
In the video the RNN is demonstrated to produce Shakespeare plays by having the network produce words one character at a time. The output of the network is fed back into the input on the next iteration.
Here's a diagram of my network:
+--------------------------------+
| |
| In: H E L L O W O R L <--+-----+
| | | | | | | | | | | | |
| V V V V V V V V V V | | Recursive feed
| +-----------------+ | |
+-> Hin ->| RNN + Softmax |-> Hout |
+-----------------+ |
| | | | | | | | | | |
Out: V V V V V V V V V V |
E L L O W O R L D ---------+
^ Character predicted by the network
I expect the network to at least do the copying bit correctly. Unfortunately my network always outputs 32 for all values (ASCII space character). I'm not sure what is causing the issue...
Please help me get my network producing poetry!
My code is here:
https://github.com/calebh/namepoet/blob/03f112ced94c3319055fbcc74a2acdb4a9b0d41c/main.py
The corpus can be replaced by a few paragraphs of Lorem Ipsum to speed up training (the network has the same bad behavior).
Sounds like it might be saturating your filters (ergo the activation function is way at the far end of the spectrum and thus has a very low gradient and gets stuck). You might want to try initializing your neurons parameters with a different method.
Also, is there a particular reason you're using GRU? In my experience LSTM units are more reliable, if a bit less efficient.
I'd try to run the code for longer time? You have batch_size = 10, sequence_size = 30 and 20 iterations, your network essentially has seen 6000 characters in total, maybe with a learning rate of 0.001, it wasn't enough to move away from your initialization.
Hence, I'd try to raise the learning rate to a very high value (e.g. 1 or 100) and see if it starts outputting different letters to confirm your implementation is somewhat correct. A network trained with such a high learning rate is usually not going to be accurate at all.
Related
i have dataset, the data set consists of : Date , ID ( the id of the event ), number_of_activities, running_sum ( the running_sum is the running sum of activities by id).
this is a part of my data :
date | id (id of the event) | number_of_activities | running_sum |
2017-01-06 | 156 | 1 | 1 |
2017-04-26 | 156 | 1 | 2 |
2017-07-04 | 156 | 2 | 4 |
2017-01-19 | 175 | 1 | 1 |
2017-03-17 | 175 | 3 | 4 |
2017-04-27 | 221 | 3 | 3 |
2017-05-05 | 221 | 7 | 10 |
2017-05-09 | 221 | 10 | 20 |
2017-05-19 | 221 | 1 | 21 |
2017-09-03 | 221 | 2 | 23 |
the goal for me is to predict the future number of activities in a given event, my question : can i train my model on all the dataset ( all the events) to predict the next one, if so how? because there are inequalities in the number of inputs ( number of rows for each event is different ), and is it possible to exploit the date data as well.
Sure you can. But alot of more information is needed, which you know yourself the best.
I guess we are talking about timeseries here as you want to predict the future.
You might want to have alook at recurrent-neural nets and LSTMs:
An Recurrent-layer takes a timeseries as input and outputs a vector, which contains the compressed information about the whole timeseries. So lets take event 156, which has 3 steps:
The event is your features, which has 3 timesteps. Each timestep has different numbers of activities (or features). To solve this, just use the maximum amount of features occuring and add a padding value (most often simply zero) so they all have the samel length. Then you have a shape, which is suitable for a recurrent neural Net (where LSTMS are currently a good choice)
Update
You said in the comments, that using padding is not option for you, let me try to convince you. LSTMs are good at situations, where the sequence length is different long. However, for this to work you also need to have longer sequences, what the model can learn its patterns from. What I want to say, when some of your sequences have only a few timesteps like 3, but you have other with 50 and more timesteps, the model might have its difficulties to predict these correct, as you have to specify, which timestep you want to use. So either, you prepare your data differently for a clear question, or you dig deeper into the topic using SequenceToSequence Learning, which is very good at computing sequences with different lenghts. For this you will need to set up a Encoder-Decoder network.
The Encoder squashs the whole sequence into one vector, whatever length it is. This one vector is compressed in a way, that it contains the information of the sequence only in one vector.
The Decoder then learns to use this vector for predicting the next outputs of the sequences. This is a known technique for machine-translation, but is suitable for any kind of sequence2sequence tasks. So I would recommend you to create such a Encoder-Decoder network, which for sure will improve your results. Have a look at this tutorial, which might help you further
I'm pretty new into Machine Learning and I was wondering if certain algorithms/models (ie. logistic regression) can handle lists as a value for their variables. Until now I've always used pretty standard datasets, where you have a couple of variables, associated values and then a classification for those set of values (view example 1). However, I now have a similar dataset but with lists for some of the variables (view example 2). Is this something logistic regression models can handle, or would I have to do some kind of feature extraction to transform this dataset into just a normal dataset like example 1?
Example 1 (normal):
+---+------+------+------+-----------------+
| | var1 | var2 | var3 | classification |
+---+------+------+------+-----------------+
| 1 | 5 | 2 | 526 | 0 |
| 2 | 6 | 1 | 686 | 0 |
| 3 | 1 | 9 | 121 | 1 |
| 4 | 3 | 11 | 99 | 0 |
+---+------+------+------+-----------------+
Example 2 (lists):
+-----+-------+--------+---------------------+-----------------+--------+
| | width | height | hlines | vlines | class |
+-----+-------+--------+---------------------+-----------------+--------+
| 1 | 115 | 280 | [125, 263, 699] | [125, 263, 699] | 1 |
| 2 | 563 | 390 | [11, 211] | [156, 253, 399] | 0 |
| 3 | 523 | 489 | [125, 255, 698] | [356] | 1 |
| 4 | 289 | 365 | [127, 698, 11, 136] | [458, 698] | 0 |
| ... | ... | ... | ... | ... | ... |
+-----+-------+--------+---------------------+-----------------+--------+
To provide some additional context on my specific problem. I'm attempting to represent drawings. Drawings have a width and height (regular variables) but drawings also have a set of horizontal and vertical lines for example (represented as a list of their coordinates on their respective axis). This is what you see in example 2. The actual dataset I'm using is even bigger, also containing variables which hold lists containing the thicknesses for each line, lists containing the extension for each line, lists containing the colors of the spaces between the lines, etc. In the end I would like to my logistic regression to pick up on what result in nice drawings. For example, if there are too many lines too close the drawing is not nice. The model should pick up itself on these 'characteristics' of what makes a nice and a bad drawing.
I didn't include these as the way this data is setup is a bit confusing to explain and if I can solve my question for the above dataset I feel like I can use the principe of this solution for the remaining dataset as well. However, if you need additional (full) details, feel free to ask!
Thanks in advance!
No, it cannot directly handle that kind of input structure. The input must be a homogeneous 2D array. What you can do, is come up with new features that capture some of the relevant information contained in the lists. For instance, for the lists that contain the coordinates of the lines along an axis (other than the actual values themselves), one could be the spacing between lines, or the total amount of lines or also some statistics such as the mean location etc.
So the way to deal with this is through feature engineering. This is in fact, something that has to be dealt with in most cases. In many ML problems, you may not only have variables which describe a unique aspect or feature of each of the data samples, but also many of them might be aggregates from other features or sample groups, which might be the only way to go if you want to consider certain data sources.
Wow, great question. I have never consider this, but when I saw other people's responses, I would have to concur, 100%. Convert the lists into a data frame and run your code on that object.
import pandas as pd
data = [["col1", "col2", "col3"], [0, 1, 2],[3, 4, 5]]
column_names = data.pop(0)
df = pd.DataFrame(data, columns=column_names)
print(df)
Result:
col1 col2 col3
0 0 1 2
1 3 4 5
You can easily do any multi regression on the fields/features of the data frame and you'll get what you need. See the link below for some ideas of how to get started.
https://pythonfordatascience.org/logistic-regression-python/
Post back if you have additional questions related to this. Or, start a new post if you have similar, but unrelated, questions.
I have a dataset where some features are numerical, some categorical, and some are strings (e.g. description). To give an example, lets say I have three features:
| Number | Type | Comment |
---------------------------------------------------------
| 1.23 | 1 | Some comment, up to 10000 characters |
| 2.34 | 2 | Different comment many words |
...
Can I have all of them as input to a multi-layer network in dl4j, where numerical and categorical would be regular input features, but string comment feature will be processed first as word-series by a simple RNN (e.g. Embedding -> LSTM)? In other words, architecture should look something like this:
"Number" "Type" "Comment"
| | |
| | Embedding
| | |
| | LSTM
| | |
Main Multi-Layer Network
|
Dense
|
...
|
Output
I think in Keras this can be achieved by Concatenate layer. Is there something like this in DL4J?
Dl4j has 99% keras import coverage. We have concatneate layers as well. Take a look at the various vertices. Whatever you can do in keras should be do able in dl4j, save for very specific cases. More here: https://deeplearning4j.org/docs/latest/deeplearning4j-nn-computationgraph You want a MergeVertex.
Let's assume that I've got patients with information about their diseases and symptoms. I want to estimate probability of P(diseasei = TRUE|symptomj = TRUE). I suppose that I should use NB classifier, but every example I've found apply Naive Bayes when there's only one disease (like predicting the probability of heart attack).
My data look like below:
patient | disease | if_disease_present | symptom
1 | d1 | TRUE | s1
2 | d1 | FALSE | s2
3 | d2 | TRUE | s1
4 | d3 | TRUE | s4
5 | d4 | FALSE | s8
...
My idea was to split data according to diseases and build the number of naive Bayesian models how many unique diseases I have in my data, but I have doubts if it's proper method.
If you want to predict the disease, don't split the data on it.
That is your target variable!
But as is, your table is not suitable for this task. You need to preprocess it, probably do some pivotization.
I am working on a project, in which we have to extract the patterns(User behavior) from the device log data. Device log contains different device actions with a timestamp like when the devices was switched on or when they was switched off.
For example:
When a person enters a room. He first switches on the light and then he
switches on the fan or Whenever the temp is less than 20 C, he switches off
the AC.
I am thinking to use Bayesian Networks to extract these patterns.
Learn Bayes Network from data (using Weka or Netica).
Arcs in Bayes Network will give the patterns/dependencies among different devices.
Is this the right approach ??
Edit: The chronological order between devices matters.
Is this the right approach?
There's many possible approaches, but here's a very simple and effective one that fits the domain:
Given the nature of the application, chronological order doesn't really matter, it doesn't matter if the Fan gets turned on before the Light e.g.
Also given that you can have e.g. a motion sensor to trigger a routine that reads the sensors, and perhaps a periodic temperature check, you can use the network below to act upon the extracted patterns (no need to complicate it further with chronological order and event tracking, we extract data to act upon, and event order in this domain isn't interesting)
For example: When a person enters a room. He first switches on the
light and then he switches on the fan or Whenever the temp is less
than 20 C, he switches off the AC.
Raw devices log might look something like this, T/F being True/False:
Person in room | Temperature | Light | Fan | AC
-----------------------------------------------
T | 20 | T | T | T
T | 19 | T | T | F
F | 18 | F | F | F
With sufficient samples you can train a model on the above, e.g. Naive bayes is not sensitive to irrelevant features/inputs, so e.g. if you look at my first raw table above that includes all the variables and try to predict AC, with sufficient data it will understand that some inputs are not very important or completely irrelevant
Or if you know how before hand what the Light, Fan, and AC depend on, e.g. we know Light isn't going to depend on Temperature, and that Fan and AC don't care if Light is turned on or not (they can operate even if the person is sleeping e.g.) so you can break it down like below:
Person in Room | Light
----------------------
T | T
F | F
Person in Room | Temperature | Fan
----------------------------------
T | 20 | T
F | 25 | F
Person in room | Temperature | AC
---------------------------------
T | 20 | T
T | 19 | F
F | 20 | F
F | 19 | F