mat kelcey

bnn. bee detection tensorflow conv net for a rasp pi on side of a hive

resemblance. trying shingling / resemblance / simhash / sketching to do some data deduping

common-crawl. playing around with the common crawl dataset

drivebot. tensorflow deep RL for driving a rover around

cartpoleplusplus. 3d cartpole gym env using bullet physics trained from pixels with tensorflow LRPG, DDPG & NAF

malmomo. tensorflow deep RL hacking on minecraft with malmo

procedural_objects. blender based random procedural object generation for bullet grasping

dither_net. a GAN experiment to generate dithers minimising pixel change between frames for an eink very slow movie player

rnn_lm. various simple RNNs trained on synthetic grammars

wikipediaPhilosophy. do all first links on wikipedia _really_ lead to philosophy?

ros-mpu6050-node. raspberry pi c++ ROS mpu6050 IMU node

collocations. bigram / trigram analysis of wikipedia; mainly mutual info

tcn_grasping. pybullet grasping with time contrastive network embeddings

ros-hc-sr04-node. raspberry pi c++ ROS hc-sr04 sonar node

common-crawl-quick-hacks. common crawl quick hack examples

lsa. latent semantic analysis experiments

median. erlang experiment in distributed median finding

doing_bayesian_data_analysis. R

snli_nn. hacking on the stanford natural language inference (SNLI) corpus (in theano)

rss-feed-experiments. machine learning experiments around a bunch of rss articles

movidius_minimal_example. minimal example of training tf conv net and deploying to movidius neural compute stick

farmbot_api. python api for bulk farm bed sweeps, image capture and plant detection

semisupervised_naive_bayes. semisupervised naive bayes

yahoo_lda_utils. some simple pre/post processing utils for yahoo lda

tgraph. mine and analyse twitter friend graph

snli_nn_tf. hacking on the stanford natural language inference (SNLI) corpus (tensorflow)

evolved_channel_selection. rather than use all 13 channels in a multi spectral image for classification can we train a model that is robust to all combos, at all resolutions, and use a genetic algorithm to choose which are the most valuable?

trending. testing out some trending algorithms, mostly written in hadoop pig

audioscrobbler-experiments. finding the shortest path from napalm death to celion dion through audioscrobbler associations

movidius_bug_reports. repro cases for bugs related to NCS api

diy_twitter_client. a simple twitter client that learns what i like to read

neural_prob_lang_model. hacky exploratory variants on NN language models

gardenhose-microslurp. bootstrap scripts for getting a micro ec2 instance piping gardenhose to s3

ros-motorhat-node. raspberry pi c++ ROS wrapper for adafruit motorhat

ensemble_net. training ensembles with jax

sip. hadoop ruby/streaming statistically improbable phrases

bullet_grasping. hacking with the bullet grasping setup

named-entity-extraction. proof of concept using NLTK for named entity extraction

mislabelled-training-data. quick experiment on an approach to correct mislabelled data

consistent_hash. afternoon hackery visualising the consistent hash method

embedding-cache. local file cache over tf.hub resnet inception feature vectors

dets_test. testing scalability aspects of erlang dets tables

text-utils. simple c++ corpus munging apps

twitter. small scale twitter experiments

statistical_synonyms. statistical approache to deriving synonyms from a large corpus

large_batch. quick hack with global batch size of 172,000

DaisyExamples. Examples for the Daisy Platform

evolved_cartpole. evolving a gym cartpole controller; directly against bytes of a tf lite microcontroller model

minimal_pybullet_on_dataflow_eg. Minimal example of running Pybullet under Google Cloud Dataflow

aliens_drum_kit. building sample embeddings so we can extract a drum kit from the audio of "aliens"

poi-dups. near duplicate detection (with ngram frequency weighting)

connected-components. test of iterative python embedded pig

twitter_follow_recommendations. simple collaborative filter recommendations to decide who i should follow based on the friends of my friends followers

dead-simple-ga. a dead simple genetic algorithm for whietpages book club