serhii.net

Semantic highlighting

This is actually really nice as idea, and as usual someone on the internet thought about this more than I did: Making Semantic Highlighting Useful - Brian Will - Medium

I somehow really like the idea of having color giving me actual semantic information about the thing I’m reading, and there are a lot of potentially cool stuffs that can be done, such as datatypes etc. It’s very connected to my old idea of creating a writing system that uses color to better and more concisely mark different letters, like the apparently defunct Dotsies but even more interesting.

Zsh autosuggestions (fish-like)

This is interesting: zsh-users/zsh-autosuggestions: Fish-like autosuggestions for zsh

Less noisy autocomplete than the default, should look similar to this:

As a side note I like the cat explanation.txt part for screenshots.

Adding numbers in Bash

integer arithmetic - How can I add numbers in a bash script - Stack Overflow

num=$((num1 + num2))
num=$(($num1 + $num2))

.. which is what I used in the updated create.sh script. FILE=_posts/$(date +%Y-%m-%d)-day$((365+$(date +%j))).markdown

Tensorflow

• TODO - why can’t tf.convert_to_tensor() convert stuff to other types (int64->float32) and I have to use tf.cast() afterwards?
• tf.in_train_phase() – both x and y have to be the same shape
• In a custom layer, compute_mask() can return a single None even if there are multiple output layers!

German

Erfahrungsmäßig

German random

The Ctrl key in Germany is “Strg”, pronounced “Steuerung”

English random

refuse - Dictionary Definition : Vocabulary.com Refuse as a verb is re-FYOOZ, as a noun it’s REF-yoss.

German

• Schmierpapier: scratch paper
• verwursten: to make into wurst.

Random

Chaostreff – Eigenbaukombinat

Tensorflow eager execution

Makes everything slower by about 2-4 times.

Tensorflow object has no attribute _keras_history

AttributeError: 'tensorflow.python.framework.ops.EagerTensor' object has no attribute '_keras_history disappears if we dont’t use eager execution inside the metric, fine if we use it inside the model. That is tf.config.experimental_run_functions_eagerly(False) inside metrics.py solves this, but model.run_eagerly=True is fine.

https://github.com/tensorflow/addons/pull/377 re output_masks and it being blocked

tf.keras vs tf.python.keras

tensorflow - What is the difference between tf.keras and tf.python.keras? - Stack Overflow

Python shell get last value

_ does the magic. Can be used in expressions too.

Python unittest

• When creating a TestCase, all vars set up in setUp should belong to the class – self.xxx
• The functions run in alphabetical order but it’s not something I should depend on

Stack / ideas

Some kind of ML language switcher that trains on my input – I write something in L1, delete, write same keystrokes on L2 => training instance. Also based on window class and time maybe?

Tensorflow ‘could not find valid device for node’

“Could not find valid device for node.” while eagerly executing. - means wrong input type.

Tensorflow

Mixing keras and tf.keras is bad karma; only the same one should be used and it should be tf.keras.

Tensorflow Keras stuff outside layers

AttributeError:’Tensor’ object has no attribute ‘_keras_history’ – inneka.com

Everything should be done by:

• Using keras.backend functions,
• Lambda layers,
• Designated keras functions with the same behavior.

When you are using tf operation – you are getting tf tensor object which doesn’t have history field. When you use keras functions you will get keras.tensors.