Classifying Handwritten Digits with TF.Learn - Machine Learning Recipes #7

[@WhirlpoolBrewer]

I love this series, please make more! I'm learning quite a bit, I can't wait to learn more! The books on this topic are often quite expensive and dig deep into math (which is fine, but often slow) and other youtubers don't really explain whats going on. You're nailing that sweet spot of teaching about whats going on, without digging too deep. I know enough to speak to whats going on, without requiring brushing up on my Calculus. Love. It.

[@rezayahya3539]

Ryan Johnson

[@robind999]

You really take us to a brand new TF world, thank you very much.

[@girishbkool]

This is the most well defined video of the Digit classifier. Awesome Job.

[@gouthamk139]

When will the next video releases .

[@XF3DeX]

This series is great! Keep it coming

[@typingcat]

Use your machine learning to predict when the next episode will come out.

[@akl6111]

Josh, You made it look so simple..atleast to get the basics. You are awesome.

[@ShawnHice]

I'm excited for the deep learning segment! Very nice tutorials I am loving it so far

[@niccolomarcon]

Awesome video! Happy to see that the thesis for my high school diploma uses the same concept :) Can't wait for the next video!

[@dmytrozelinskyi4824]

Hey. Don`t u think it`s time to make new cool video about ML ? Thx for last 7n.

[@juanok2775]

These are the best videos, we need more!!

[@JacquesvanWyk]

Wow a whole new world opened.This is great stuff cant wait to learn more.

[@AmCanTech]

I am loving this series. was working with caffe for a bit, but need to start myself with TF

[@yurielzhang5884]

I can't wait to see the next episode, it's awesome, thank you very much

[@shmigelsky]

Awesome, looking forward to the deep learning episode. Would be interesting to see learning times for different approaches and hardware. G

[@RobertShaverOfAustin]

Some people put a slash through the zero and the seven character when they write it. How would you incorporate those alternative forms into this digit recognizer? Does the training set contain data about what each character is actually (a designator)? If that is true it seems like you could include the alternative digit forms with the correct designator. Then both forms would contribute to the weights for those two other characters, zero and seven, right?

[@afederdk]

You simply would provide a training data set containing a sufficient number of examples of those alternative forms. The classifier is as good as the data it is trained on.

[@parampalsingh3339]

Continue the series pleaseeeee....I'm lovin it

[@maplicant232]

Thanks for doing this! I have a question though. Would the classifier get better results if you had 10 extra weights: biases on the outputs? It seems like it would, but I don't know how much of a difference it would make.

[@shahroozgovahi674]

Thank you for your effort in creating quality material in ML. The method you use for explaining is amazing. Is there any more videos beside these 7?

[@SaaranshMehta]

I'm a fan of your videos, thanks for the lucid explanation

[@_akuma06]

I want MOAR ! Thanks for the series, it was really helpful :)

[@dc5]

Time to go DEEP :) Great videos... Thanks for sharing Josh!

[@yousufazad6914]

please continue making these awesome videos!

[@sallerc]

Great series, hope you do more episodes.

[@yosaikan]

Thank you Joch for this very clear presenation :)

[@chiranjib-konwar]

why just 7 videos only. please make more videos. you are just awesome. thanks a lot.

[@Tzvetkov]

So weights are like pixel intensities of the digits learned by the model?

[@kamikaz1k]

Anyone having problems with running the classifier.predict method? The evaluation method works (it gives me an accuracy score) but the predict method always throws an InvalidArgumentError saying that "shape in shape_and_slice spec [1,10] does not match the shape stored in checkpoint: [784,10]"

[@chinamatt]

I get the same error...

[@chinamatt]

I found this works too: pred = clf.predict(np.array([test_data[0]]), as_iterable=False) print("Predicted %d, Label: %d" % (pred, test_labels[0]))

[@robertlam88]

Thanks I ran into the same error and the above works.

[@thinkcreativewithsakshi259]

I am getting the same error dude. Please respond how you solved it! Thank you.

[@adambrown234]

This was a really nice video, it helped me a lot. Thanks! When is the next episode coming up? I'd love to see it.

[@juansantiagocuadra3672]

What tools are you using to create the animations? great videos, Thank You.

[@Taha-jj1kr]

Beautiful ...just beautiful explanation !!

[@Xenotropic]

If you get an error using a newer version of TensorFlow when trying to predict from the test data try the following: replace - print ("Predicted %d, Label: %d" % (classifier.predict(test_data[0]), test_labels[0])) with - print ("Predicted %d, Label: %d" % (classifier.predict(np.array([test_data[0]], dtype=float), as_iterable=False))

[@nguyenthanhdong]

Thank you

[@mpik97]

Looking forward to the next video!

[@jonathanmv]

Im getting an error when trying to visualize the weights *weights = classifier.weights_* throws an _AttributeError: 'LinearClassifier' object has no attribute 'weights_' error. Does anyone have a clue about what's happening or how to the the weights for the classifier?

[@yording54cala]

hi, i have the same error, Could you solve it?

[@adeeb12321]

me too

[@isco2252]

Me too. I've checked the library. No attribute weights_ in it.

[@jonasschmeiss5487]

short: replace "weights = classifier.weights_" with for var in classifier.get_variable_names(): print("var:", var, "=", classifier.get_variable_value(var)) weights = classifier.get_variable_value("linear//weight/d/linear//weight/part_0/Ftrl_1")

[@danielrichards3159]

I found the answer, just DELETE THE LINE (seriously, it works)

[@bantuthomas]

Hi!! My Anaconda installation does not have the tensorflow module. Neither could I install it through conda. Similar problem with IDLE and PyCharm. I get the error message: Could not find a version that satisfies the requirement tensorflow (from versions: ) No matching distribution found for tensorflow I am stuck without being able to install tensorflow. Is there any IDE the comes with tensorflow?

[@splashgi]

Love the series. When will next video be released?

[@matheoland]

Good episode. Thanks !

[@nguyenthanhdong]

How do I predict for a single image in my computer?

[@user-jk3dq2fp7n]

I can't seem to find the Docker quickstart terminal on my Mac. I am using Version 17.03.0-ce-mac1.

[@NathboGames]

Hi once classifier works well what is the next step to build a program to use the results to return something to user? Ie I would want to read digits and maybe a + - = so I would classify the image as 1+2= and my program should return the result of the math operation, so 3. What are the ways to do that?

[@Realxar]

The error was: SyntaxError: Unexpected token < in JSON at position 4

[@eddalexander6030]

Yes getting same errorr too if anyone knows how to resolve

[@DanielMarrable]

The "Visualizating MNIST" link is missing a space between the link and "(this blog is outstanding)" and as such the link gets garbled and 404s. Great video as always!!

[@eliesaad]

What happened? why did the series stop? when's the next video?

[@nikitatank6672]

Is docker installation necessary for this tutorial ?? If no. I am getting "AttributeError: 'LinearClassifier' object has no attribute 'weights' " error. Not able to print predicted and actual label. May i have the solution please

[@RAJATTHEPAGAL]

We need more videos . :-D ... Josh Gordon awesome introduction from you sir.

[@Johnny5490]

What if the numbers are not centered, isn't that an easy way to beat this? Like if the one was left of the center pixel

[@jiteshmohite1069]

Great videos on machine learning. also it would be nice if you guys come up with some more videos with much deeper knowledge.

[@thinkcreativewithsakshi259]

Helli Sir, I ran the above source code. Well, it is giving me an error at the "Classify few examples" point where we are meant to predict the actual number. Can you please help. Thank you.

[@bmusician08]

Now that TF2 is out all these Episodes are obsolete. Does any know where to find updates to them?

[@tjhannover3069]

is their a diffrence between skit learn and tensflow?

[@abishekkumar316]

Unable to start the quick start terminal for docker

[@miguelpereira9613]

"I’d like to spend a lot more time on this next episode, if there’s interest?" When? :D Love your videos and your links btw!

[@yuval9064]

when i run this code i get an error with classifier.predict help?

[@SirajRaval]

Hell yeah! If you guys like machine learning check out my new ML series on my channel.

[@patyarakawa]

I have a doubt. In the examples you use python. Have you tried some functional language, such as elixir, for example? I am interested in deepening machine learning but I still do not know which language is the best. Thanks for the videos, by the way!

[@JoeWong81]

Thank you for these videos buddy!!!

[@conradosanchez7332]

Anyone know if this is with multilayer perceptron or convolutional neural networks?

[@actechforlife]

when is the episode on deep learning going to start.... kindly post link if already.. thanks..

[@narendranathp]

Are more episodes coming in?

[@OussamaHaff]

Can I use Tensorflow to do OCR on Android ?

[@mitrathakker]

Hey Josh It was a great series so far and expecting more such videos But there was this one thing that constantly pricked me during the whole playlist and that's the order of the videos It is reverse chronological and for someone like me who started watching the playlist once it already had 7 videos, it got really cumbersome (and sometimes confusing) to actually go to the previous video after each video I saw Is it possible to get this (and most others, too) playlist ordered in a chronological fashion (or maybe get a "Reverse Playlist" functionality on KZbin as that'd work too :P) Cheers

[@Skandawin78]

is this the last video in this series ? :(

[@IagobaApellaniz]

the order of the playlists is reversed

[@wessidemd]

This should be the top comment. The reverse order makes it difficult to watch them sequentially.

[@priyankamaurya8175]

if i want to input an image then how it recognize that image

[@sheresh]

what are weights in ML ?

[@sudharsanamtk]

I am planning to develop a chat bot in non english language? where to start ? is TF an option or anything ? please suggest

[@lakshmichandubathini6354]

do docker doesn't work in windows 8.1?

[@huynguyen-cy8ig]

When the next video come out, I've been waiting for too long!

[@AlexanderCastroRomero]

I am a teacher and this video is so userfull :)

[@madhuvamsimachavarapu5267]

is this the last episode?

[@yixunnnn]

Hi guys! Does anyone knows what version to use as of this moment? Thanks!

[@yixunnnn]

Is it 1.4.0-rc0-py3?

[@joostavontuur2144]

i just installed it all

[@chrispeng5269]

why the episode is not updated?

[@xingyubian5654]

I am getting Error response from daemon: driver failed programming external connectivity on endpoint objective_volhard

[@ashii007]

I ran into env issues with ep7 notebook but I got the gist of it.

[@saikatsutradhar1138]

when are you starting classes again? eagerly waiting

[@Endlessvoidsutidos]

how do you download the ipynb file from git hub ?

[@ViventhraaRao]

Anyone got this issue when trying to execute the code? "AttributeError: 'module' object has no attribute 'datasets'"

[@AleksandrKrasnovsoprof]

Probably wrong tensorflow version.

[@cjjb]

Really liked these. Pity there's no more :(

[@morwar_]

very good, thanks for the explanation

[@russellthurman5509]

Under the "Fit Linear Classifier" sections, I don't get this return: LinearClassifier(). This is what I get: LinearClassifier(params={'head': , 'feature_columns': [_RealValuedColumn(column_name='', dimension=784, default_value=None, dtype=tf.float32, normalizer=None)], 'optimizer': None, 'gradient_clip_norm': None, 'joint_weights': False}). Then the evaluation gives me 0.0988.

[@rahulagarwal7303]

we need more videos. plz complete the series.

[@DilanGamachchige]

Nice interesting one.. waiting for the next one.. thanks.. :-)

[@danielassayag1484]

libraries used arent valid anymore on tensorflow 2. though some of it is deprecated its still on ai.google/education/.. google can"t do better than this?

[@yccheng3864]

Hi i encountered issue when I run the jupyter notebook: --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in 7 get_ipython().run_line_magic('matplotlib', 'inline') 8 import tensorflow as tf ----> 9 learn = tf.contrib.learn 10 tf.logging.set_verbosity(tf.logging.ERROR) AttributeError: module 'tensorflow' has no attribute 'contrib' I just found out this is about 3 years ago, so is that the "contrib" no longer available?

[@yccheng3864]

oh i solved it, thank you

[@RasmusKwerther]

@@yccheng3864 would you say how? :)

[@ahmedalhisaie7698]

thanks for this great tutorials

[@pacozaaskr14]

Thank you Josh.

[@jugadexpert4687]

sir google finance ke event code kya hai

[@user-ny8xt1ui3h]

Wonderful Series of ML! Someone recommends me another one? please

[@aakashawase1948]

when are you going to release next video

[@moinshaikh1333]

The GitHub link died please send the TF examples proper link

[@shikhanshu]

where is #8??

[@nkf31]

Hope text mining will be covered in one episode!

[@gideonamanimarress7441]

awesome video!! THANK YOU

[@cashphattichaddi]

Beautiful !!

[@rabanaahoura7593]

Great ! Thanks , Can i ask you about WIZARDPKG for other solution ?

[@ZIAURREHMAN1991]

please comeback soon

[@juliozarate5]

excellent all videos

[@krasin]

The link to Docker images leads to the MNIST tutorial.

[@akmonra]

The hell? There are none after this?

[@tiagoguimaraes5796]

+1 for docker!