markdown stringlengths 0 1.02M | code stringlengths 0 832k | output stringlengths 0 1.02M | license stringlengths 3 36 | path stringlengths 6 265 | repo_name stringlengths 6 127 |
|---|---|---|---|---|---|
度分布 | from collections import defaultdict
import numpy as np
def plotDegreeDistribution(G):
degs = defaultdict(int)
for i in G.degree().values(): degs[i]+=1
items = sorted ( degs.items () )
x, y = np.array(items).T
y_sum = np.sum(y)
y = [float(i)/y_sum for i in y]
plt.plot(x, y, 'b-o')
plt.xs... | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
网络科学理论简介****** 网络科学：分析网络结构******王成军 wangchengjun@nju.edu.cn计算传播网 http://computational-communication.com 规则网络 | import networkx as nx
import matplotlib.pyplot as plt
RG = nx.random_graphs.random_regular_graph(3,200) #生成包含200个节点、每个节点有3个邻居的规则图RG
pos = nx.spectral_layout(RG) #定义一个布局，此处采用了spectral布局方式，后变还会介绍其它布局方式，注意图形上的区别
nx.draw(RG,pos,with_labels=False,node_size = 30) #绘制规则图的图形，with_labels决定节点是非带标签（编号），node_size是节点的直径
... | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
ER随机网络 | import networkx as nx
import matplotlib.pyplot as plt
ER = nx.random_graphs.erdos_renyi_graph(200,0.05) #生成包含20个节点、以概率0.2连接的随机图
pos = nx.shell_layout(ER) #定义一个布局，此处采用了shell布局方式
nx.draw(ER,pos,with_labels=False,node_size = 30)
plt.show()
plotDegreeDistribution(ER) | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
小世界网络 | import networkx as nx
import matplotlib.pyplot as plt
WS = nx.random_graphs.watts_strogatz_graph(200,4,0.3) #生成包含200个节点、每个节点4个近邻、随机化重连概率为0.3的小世界网络
pos = nx.circular_layout(WS) #定义一个布局，此处采用了circular布局方式
nx.draw(WS,pos,with_labels=False,node_size = 30) #绘制图形
plt.show()
plotDegreeDistribution(WS)
nx.diameter(WS... | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
BA网络 | import networkx as nx
import matplotlib.pyplot as plt
BA= nx.random_graphs.barabasi_albert_graph(200,2) #生成n=20、m=1的BA无标度网络
pos = nx.spring_layout(BA) #定义一个布局，此处采用了spring布局方式
nx.draw(BA,pos,with_labels=False,node_size = 30) #绘制图形
plt.show()
plotDegreeDistribution(BA)
BA= nx.random_graphs.barabasi_albert_grap... | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
作业：- 阅读 Barabasi (1999) Internet Diameter of the world wide web.Nature.401- 绘制www网络的出度分布、入度分布- 使用BA模型生成节点数为N、幂指数为$\gamma$的网络- 计算平均路径长度d与节点数量的关系 | Ns = [i*10 for i in [1, 10, 100, 1000]]
ds = []
for N in Ns:
print N
BA= nx.random_graphs.barabasi_albert_graph(N,2)
d = nx.average_shortest_path_length(BA)
ds.append(d)
plt.plot(Ns, ds, 'r-o')
plt.xlabel('$N$', fontsize = 20)
plt.ylabel('$<d>$', fontsize = 20)
plt.xscale('log')
plt.show() | _____no_output_____ | MIT | code/17.networkx.ipynb | nju-teaching/computational-communication |
Now we have to put together the different datasets. | import pandas as pd
files_df = pd.read_pickle("data/clean/files_df.pkl") | _____no_output_____ | MIT | S02 - Data Wrangling/HCKT02 - Data Wrangling/instructor_solution/5.putting_all_together.ipynb | jtiagosg/batch3-students |
We remove from the dataframe those rows whose origin is the website (these rows have `WEBSITE` in all the values) and those that come from the API (these rows have `API` in all the values). | files_df.shape
website_ids = files_df[files_df.tierafterorder.isin(['WEBSITE'])].index
api_ids = files_df[files_df.tierafterorder.isin(['API'])].index
files_df = files_df[-files_df.tierafterorder.isin(['WEBSITE', 'API'])]
files_df.shape
files_df.tierafterorder.value_counts()
scraped_df = pd.read_pickle('data/clean/scra... | _____no_output_____ | MIT | S02 - Data Wrangling/HCKT02 - Data Wrangling/instructor_solution/5.putting_all_together.ipynb | jtiagosg/batch3-students |
We concat the dataframes that have different ids. | train_df = pd.concat(
[
pd.concat([files_df, api_df], sort=True),
scraped_df
],
sort=True
)
train_df.shape | _____no_output_____ | MIT | S02 - Data Wrangling/HCKT02 - Data Wrangling/instructor_solution/5.putting_all_together.ipynb | jtiagosg/batch3-students |
Now we join the files that share an index. | train_df = train_df.drop(columns=['returned', 'storeid']).join(targets_df).join(storeid_df)
train_df.shape
train_df.to_pickle('data/clean/train_df_merged.pkl') | _____no_output_____ | MIT | S02 - Data Wrangling/HCKT02 - Data Wrangling/instructor_solution/5.putting_all_together.ipynb | jtiagosg/batch3-students |
Recommendations with IBMIn this notebook, you will be putting your recommendation skills to use on real data from the IBM Watson Studio platform. You may either submit your notebook through the workspace here, or you may work from your local machine and submit through the next page. Either way assure that your code p... | %autosave 180
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import project_tests as t
import pickle
import seaborn as sns
import re
import nltk
from nltk.corpus import stopwords
from nltk.stem.wordnet import WordNetLemmatizer
from nltk.tokenize import word_tokenize
from sklearn.feature_extract... | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Part I : Exploratory Data AnalysisUse the dictionary and cells below to provide some insight into the descriptive statistics of the data.`1.` What is the distribution of how many articles a user interacts with in the dataset? Provide a visual and descriptive statistics to assist with giving a look at the number of ti... | # make a groupby instance and count how many articles were read by each user
email_grouped_df = df.groupby('email')
num_article_email = email_grouped_df['article_id'].count()
print("Mean # article :",num_article_email.mean())
print("Quantile 0.25 , 0.5, 0.75: " , num_article_email.quantile(0.25), num_article_email.quan... | 50% of individuals interact with 3.0 number of articles or fewer.
The maximum number of user-article interactions by any 1 user is364:
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`2.` Explore and remove duplicate articles from the **df_content** dataframe. | # Find and explore duplicate articles
df_content.head()
check_dupl_df_1 = df_content[df_content.duplicated(['article_id'])]
check_dupl_df_1
# Remove any rows that have the same article_id - only keep the first
df_content.drop_duplicates(subset='article_id', keep='first', inplace=True)
check_dupl_df_1 = df_content[df_co... | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`3.` Use the cells below to find:**a.** The number of unique articles that have an interaction with a user. **b.** The number of unique articles in the dataset (whether they have any interactions or not).**c.** The number of unique users in the dataset. (excluding null values) **d.** The number of user-article interac... | unique_articles = len(df.article_id.unique()) # The number of unique articles that have at least one interaction
total_articles = len(df_content.article_id.unique()) # The number of unique articles on the IBM platform
df_email_na_dropped = df.dropna(subset=['email'])
unique_users = len(df_email_na_dropped.email.unique(... | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`4.` Use the cells below to find the most viewed **article_id**, as well as how often it was viewed. After talking to the company leaders, the `email_mapper` function was deemed a reasonable way to map users to ids. There were a small number of null values, and it was found that all of these null values likely belong... | # df_content.head(3)
article_id_grouped_df = df.groupby('article_id')
print(article_id_grouped_df['email'].count().sort_values(ascending=False).index[0])
print(article_id_grouped_df['email'].count().sort_values(ascending=False).values[0])
most_viewed_article_id = '1429.0'# The most viewed article in the dataset as a st... | It looks like you have everything right here! Nice job!
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Part II: Rank-Based RecommendationsUnlike in the earlier lessons, we don't actually have ratings for whether a user liked an article or not. We only know that a user has interacted with an article. In these cases, the popularity of an article can really only be based on how often an article was interacted with.`1.` ... | def get_top_articles(n, df=df):
'''
INPUT:
n - (int) the number of top articles to return
df - (pandas dataframe) df as defined at the top of the notebook
OUTPUT:
top_articles - (list) A list of the top 'n' article titles
'''
article_id_grouped_df = df.groupby(['title'])
... | Your top_5 looks like the solution list! Nice job.
Your top_10 looks like the solution list! Nice job.
Your top_20 looks like the solution list! Nice job.
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Part III: User-User Based Collaborative Filtering`1.` Use the function below to reformat the **df** dataframe to be shaped with users as the rows and articles as the columns. * Each **user** should only appear in each **row** once.* Each **article** should only show up in one **column**. * **If a user has interacted... | # create the user-article matrix with 1's and 0's
def create_user_item_matrix(df):
'''
INPUT:
df - pandas dataframe with article_id, title, user_id columns
OUTPUT:
user_item - user item matrix
Description:
Return a matrix with user ids as rows and article ids on the columns with ... | You have passed our quick tests! Please proceed!
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`2.` Complete the function below which should take a user_id and provide an ordered list of the most similar users to that user (from most similar to least similar). The returned result should not contain the provided user_id, as we know that each user is similar to him/herself. Because the results for each user here ... | def find_similar_users(user_id, user_item=user_item):
'''
INPUT:
user_id - (int) a user_id
user_item - (pandas dataframe) matrix of users by articles:
1's when a user has interacted with an article, 0 otherwise
OUTPUT:
similar_users - (list) an ordered list where the closes... | The 10 most similar users to user 1 are: [3933, 23, 3782, 203, 4459, 3870, 131, 4201, 46, 3697]
The 5 most similar users to user 3933 are: [1, 3782, 23, 203, 4459]
The 3 most similar users to user 46 are: [4201, 3782, 23]
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`3.` Now that you have a function that provides the most similar users to each user, you will want to use these users to find articles you can recommend. Complete the functions below to return the articles you would recommend to each user. | def get_article_names(article_ids, df=df):
'''
INPUT:
article_ids - (list) a list of article ids
df - (pandas dataframe) df as defined at the top of the notebook
OUTPUT:
article_names - (list) a list of article names associated with the list of article ids
(this is iden... | If this is all you see, you passed all of our tests! Nice job!
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`4.` Now we are going to improve the consistency of the **user_user_recs** function from above. * Instead of arbitrarily choosing when we obtain users who are all the same closeness to a given user - choose the users that have the most total article interactions before choosing those with fewer article interactions.* ... | def get_top_sorted_users(user_id, df=df, user_item=user_item):
'''
INPUT:
user_id - (int)
df - (pandas dataframe) df as defined at the top of the notebook
user_item - (pandas dataframe) matrix of users by articles:
1's when a user has interacted with an article, 0 otherwise
... | The top 10 recommendations for user 20 are the following article ids:
['1162.0', '1351.0', '1164.0', '491.0', '1186.0', '14.0', '1429.0', '162.0', '939.0', '813.0']
The top 10 recommendations for user 20 are the following article names:
['analyze energy consumption in buildings', 'model bike sharing data with spss', '... | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`5.` Use your functions from above to correctly fill in the solutions to the dictionary below. Then test your dictionary against the solution. Provide the code you need to answer each following the comments below. | ### Tests with a dictionary of results
neighbor_df_1 = get_top_sorted_users(1, df=df, user_item=user_item)
neighbor_df_131 = get_top_sorted_users(131, df=df, user_item=user_item)
user1_most_sim = neighbor_df_1.neighbor_id[0].item()# Find the user that is most similar to user 1
user131_10th_sim = neighbor_df_131.neigh... | This all looks good! Nice job!
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`6.` If we were given a new user, which of the above functions would you be able to use to make recommendations? Explain. Can you think of a better way we might make recommendations? Use the cell below to explain a better method for new users. Rank based recommendation is suitable for a new user because it only depe... | new_user = '0.0'
# top_10 = get_top_articles(10)
top_10 = get_top_article_ids(10, df=df)
# What would your recommendations be for this new user '0.0'? As a new user, they have no observed articles.
# Provide a list of the top 10 article ids you would give to
top_10 = list(map(str, top_10))
new_user_recs = top_10# You... | That's right! Nice job!
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Part IV: Content Based Recommendations (EXTRA - NOT REQUIRED)Another method we might use to make recommendations is to perform a ranking of the highest ranked articles associated with some term. You might consider content to be the **doc_body**, **doc_description**, or **doc_full_name**. There isn't one way to creat... | def make_content_recs(article_id, df_content, df, m=10):
'''
INPUT:
article_id = (int) a article id in df_content
m - (int) the number of recommendations you want for the user
df_content - (pandas dataframe) df_content as defined at the top of the notebook
df - (pandas dataframe) df as defined... | [730, 194, 53, 470, 1005, 980, 423, 266, 681, 670]
************************************************************
['Developing for the IBM Streaming Analytics service', 'Data science for real-time streaming analytics', 'Introducing Streams Designer', 'What’s new in the Streaming Analytics service on Bluemix', 'Real-time ... | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`2.` Now that you have put together your content-based recommendation system, use the cell below to write a summary explaining how your content based recommender works. Do you see any possible improvements that could be made to your function? Is there anything novel about your content based recommender? This part is ... | def make_content_recs_2(article_id, df_content, df, m=10):
'''
INPUT:
article_id = (int) a article id in df_content
m - (int) the number of recommendations you want for the user
df_content - (pandas dataframe) df_content as defined at the top of the notebook
df - (pandas dataframe) df as defin... | input only user_id or article_id
[] []
[384, 805, 48, 662, 809, 161, 893, 686, 723, 655] ['Continuous Learning on Watson', 'Machine Learning for everyone', 'Data Science Experience Documentation', 'Build Deep Learning Architectures With Neural Network Modeler', 'Use the Machine Learning Library', 'Use the Machine Learn... | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Part V: Matrix FactorizationIn this part of the notebook, you will build use matrix factorization to make article recommendations to the users on the IBM Watson Studio platform.`1.` You should have already created a **user_item** matrix above in **question 1** of **Part III** above. This first question here will just... | # Load the matrix here
user_item_matrix = pd.read_pickle('user_item_matrix.p')
# quick look at the matrix
user_item_matrix.head()
user_item_matrix.shape
user_item_matrix.to_numpy() | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`2.` In this situation, you can use Singular Value Decomposition from [numpy](https://docs.scipy.org/doc/numpy-1.14.0/reference/generated/numpy.linalg.svd.html) on the user-item matrix. Use the cell to perform SVD, and explain why this is different than in the lesson. | # Perform SVD on the User-Item Matrix Here
u, s, vt = np.linalg.svd(user_item_matrix)# use the built in to get the three matrices
s.shape, u.shape, vt.shape | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Here, the user-item matrix passed in linalg.svd has no missing values. All elements in the matrix are 0 or 1. In the previous lesson, there were a lot of null cells in the matrix. It was not able to be passed in. `3.` Now for the tricky part, how do we choose the number of latent features to use? Running the below cel... | num_latent_feats = np.arange(10,700+10,20)
sum_errs = []
for k in num_latent_feats:
# restructure with k latent features
s_new, u_new, vt_new = np.diag(s[:k]), u[:, :k], vt[:k, :]
# take dot product
user_item_est = np.around(np.dot(np.dot(u_new, s_new), vt_new))
# compute error for each p... | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`4.` From the above, we can't really be sure how many features to use, because simply having a better way to predict the 1's and 0's of the matrix doesn't exactly give us an indication of if we are able to make good recommendations. Instead, we might split our dataset into a training and test set of data, as shown in ... | df_train = df.head(40000)
df_test = df.tail(5993)
def create_test_and_train_user_item(df_train, df_test):
'''
INPUT:
df_train - training dataframe
df_test - test dataframe
OUTPUT:
user_item_train - a user-item matrix of the training dataframe
(unique users for each r... | Awesome job! That's right! All of the test movies are in the training data, but there are only 20 test users that were also in the training set. All of the other users that are in the test set we have no data on. Therefore, we cannot make predictions for these users using SVD.
| MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`5.` Now use the **user_item_train** dataset from above to find U, S, and V transpose using SVD. Then find the subset of rows in the **user_item_test** dataset that you can predict using this matrix decomposition with different numbers of latent features to see how many features makes sense to keep based on the accurac... | # fit SVD on the user_item_train matrix
u_train, s_train, vt_train = np.linalg.svd(user_item_train) # fit svd similar to above then use the cells below
# Use these cells to see how well you can use the training
# decomposition to predict on test data
both_rows = user_item_train.index.isin(test_idx)
rows_mask = np.inte... | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
`6.` Use the cell below to comment on the results you found in the previous question. Given the circumstances of your results, discuss what you might do to determine if the recommendations you make with any of the above recommendation systems are an improvement to how users currently find articles? 1. Brief summary of... | print("# of row / sum of readings / sum of difference")
for i in range(20):
print(i, " ",user_item_test.iloc[i].abs().sum(), " ",diffs_test.iloc[i].abs().sum()) | # of row / sum of readings / sum of difference
0 2.0 12.0
1 7.0 31.0
2 5.0 16.0
3 5.0 6.0
4 1.0 5.0
5 32.0 48.0
6 3.0 36.0
7 55.0 65.0
8 1.0 2.0
9 26.0 26.0
10 8.0 24.0
11 1.0 2.0
12 1.0 2.0... | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
ExtrasUsing your workbook, you could now save your recommendations for each user, develop a class to make new predictions and update your results, and make a flask app to deploy your results. These tasks are beyond what is required for this project. However, from what you learned in the lessons, you certainly capabl... | from subprocess import call
call(['python', '-m', 'nbconvert', 'Recommendations_with_IBM.ipynb']) | _____no_output_____ | MIT | notebook/Recommendations_with_IBM.ipynb | dalpengholic/Udacity_Recommendations_with_IBM |
Generalised RegressionIn this notebook, we will build a generalised regression model on the **electricity consumption** dataset. The dataset contains two variables - year and electricity consumption. | #importing libraries
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression
from sklearn.pipeline import Pipeline
from sklearn import metrics
#fetching data
elec_cons = pd.read_csv("total-electricity-consumption-us.csv",... | [1, 2, 3]
[0.84237474021761372, 0.99088967445535958, 0.9979789881969624]
[0.81651704638268097, 0.98760805026754717, 0.99848974839924587]
| MIT | Section 16/AdvanceReg/Teclov_generalised_regression.ipynb | ashokjohn/ML_RealWorld |
Import library | # !pip install --upgrade tables
# !pip install eli5
# !pip install xgboost
# !pip install hyperopt
import pandas as pd
import numpy as np
import xgboost as xgb
from sklearn.metrics import mean_absolute_error as mae
from sklearn.model_selection import cross_val_score
from hyperopt import hp, fmin, tpe, STATUS_OK
impo... | _____no_output_____ | MIT | matrix_two/day5.ipynb | jedrzejd/dw_matrix_car |
Feature Engineering | SUFFIX_CAT = '__cat'
for feat in df.columns:
if isinstance(df[feat][0], list):
continue
factorized_values = df[feat].factorize()[0]
if SUFFIX_CAT in feat:
continue
df[feat + SUFFIX_CAT] = factorized_values
df['param_pojemność-skokowa'] = df['param_pojemność-skokowa'].map(lambda x: -1 ... | Training with params:
{'colsample_bytree': 0.6000000000000001, 'learning_rate': 0.15000000000000002, 'max_depth': 13, 'n_estimators': 100, 'objective': 'reg:squarederror', 'seed': 0, 'subsample': 0.7000000000000001}
8021.26782298684
Training with params:
{'colsample_bytree': 0.8500000000000001, 'learning_rate': 0.2, 'm... | MIT | matrix_two/day5.ipynb | jedrzejd/dw_matrix_car |
Best Config XGBoost | feats = ['param_napęd__cat', 'param_rok-produkcji', 'param_stan__cat', 'param_skrzynia-biegów__cat', 'param_faktura-vat__cat', 'param_moc', 'param_marka-pojazdu__cat', 'feature_kamera-cofania__cat', 'param_typ__cat', 'param_pojemność-skokowa', 'seller_name__cat', 'feature_wspomaganie-kierownicy__cat', 'param_model-poja... | _____no_output_____ | MIT | matrix_two/day5.ipynb | jedrzejd/dw_matrix_car |
Obtaining Statistics of the RoomNav dataset 1. Average Geodesic Distances2. Histogram of distances vs episodes3. Average of top-down maps4. Lenght of oracle | import habitat
import numpy as np
import random
%matplotlib inline
import matplotlib.pyplot as plt
splits = ['train']
data_path = '../data/datasets/roomnav/mp3d/v1/{split}/{split}.json.gz'
for split in splits:
avg_gd = 0
avg_ed = 0
min_gd = 10000000000
max_gd = 0
min_ed = 10000000000
m... | _____no_output_____ | MIT | notebooks/dataset_statistics.ipynb | medhini/habitat-api |
Validataion set | embs_model = learn.model.eval()
embs_model.outputEmbs = True
valid_embs, _ = embs_from_model(embs_model, dls.valid)
dists, inds = get_nearest(valid_embs, do_chunk(valid_embs))
valid_df=train_df[train_df.is_valid==True].copy().reset_index()
valid_df = add_target_groups(valid_df)
pairs = sorted_pairs(dists, inds)[:len(va... | _____no_output_____ | MIT | SBert.ipynb | slawekslex/shopee |
Create a regulus file from a csv knn is the size of the neighborhood. The default is 100, which is usually sufficient | import regulus
gauss4 = regulus.from_csv('gauss4', knn=8)
regulus.save(gauss4, filename='gauss4') | _____no_output_____ | BSD-3-Clause | examples/0-gauss_to_regulus.ipynb | yarden-livnat/ipyregulus |
TSG081 - Get namespaces (Kubernetes)====================================Description-----------Get the kubernetes namespacesSteps----- Common functionsDefine helper functions used in this notebook. | # Define `run` function for transient fault handling, suggestions on error, and scrolling updates on Windows
import sys
import os
import re
import json
import platform
import shlex
import shutil
import datetime
from subprocess import Popen, PIPE
from IPython.display import Markdown
retry_hints = {} # Output in stderr... | _____no_output_____ | MIT | Big-Data-Clusters/CU3/Public/content/monitor-k8s/tsg081-get-kubernetes-namespaces.ipynb | gantz-at-incomm/tigertoolbox |
Show the Kubernetes namespaces | run('kubectl get namespace') | _____no_output_____ | MIT | Big-Data-Clusters/CU3/Public/content/monitor-k8s/tsg081-get-kubernetes-namespaces.ipynb | gantz-at-incomm/tigertoolbox |
Show the Kubernetes namespaces with labelsKubernetes namespaces containing a SQL Server Big Data Cluster have thelabel ‘MSSQL\_CLUSTER’ | run('kubectl get namespaces -o custom-columns=NAME:.metadata.name,STATUS:.status.phase,LABELS:.metadata.labels')
print('Notebook execution complete.') | _____no_output_____ | MIT | Big-Data-Clusters/CU3/Public/content/monitor-k8s/tsg081-get-kubernetes-namespaces.ipynb | gantz-at-incomm/tigertoolbox |
Introduction to RThis introduction to the R language aims at understanding how to represent and manipulate data objects as commonly found in *data science*, to provide basic summary statistic and to build relevant graphical representation of the data. **Important notice:** Only base commands are discussed here, not th... | library(ggplot2)
theme_set(theme_minimal()) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Note that you need to load the `ggplot2` package only once, at the start of your R session. Getting started VariablesThere are fundamentally two kind of data structures in statistics-oriented programming languages: numbers and strings. Numbers can be integers or real numbers and they are used to represent values obser... | x <- c(1, 3, 2, 5, 4) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Note that the symbol `<-` stands for the recommended assignment operator, yet it is possible to use `=` to assign some quantity to a given variable, which appears on the left hand side of the above expression. Also, the series of values is reported between round brackets, and each values is separated by a comma. From n... | length(x)
typeof(x) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
It should be noted that `x` contains values stored as real numbers (`double`) while they may just be stored as integers. It is however possible to ask R to use truly integer values: | x <- c(1L, 3L, 2L, 5L, 4L)
typeof(x) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
The distinction between 32 bits integers and reals will not be that important in common analysis tasks, but it is important to keep in mind that it is sometimes useful to check whether data are represented as expected, especially in the case of categorical variables, also called 'factor' in R parlance (more on this lat... | x <- c(c(1, 2, 3), c(4, 5, 6), 7, 8) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
In passing, note that since we use the same name for our newly created variable, `x`, the old content referenced in `x` (1, 3, 2, 5, 4) is definitively lost. Once you have a vector of values, you can access each item by providing the (one-based) index of the item(s), e.g.: | x[1]
x[3]
x[c(1,3)]
x[1:3] | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
A convenient shorhand notation for regular sequence of integers is `start:end`, where `start` is the starting value and `end` is the last value (both included). Hence, `c(1,2,3,4)` is the same as `1:4`. This is useful when one wants to preview the first 3 or 5 values in a vector, for example. A more general function to... | seq(1, 10)
seq(1, 10, by = 2)
seq(0, 10, length = 5) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Updating content of a vector can be done directly by assigning a new value to one of the item: | x[3] <- NA | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
In the above statement, the third item has been assigned a missing value, which is coded as `NA` ('not available') in R. Again, there is no way to go back to the previous state of the variable, so be careful when updating the content of a variable.The presence of missing data is important to check before engaging into ... | is.na(x)
which(is.na(x)) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Notice that many functions like `is.na`, or `which`, act in a vectorized way, meaning that you don't have to iterate manually over each item in the vector. Moreover, function calls can be nested one into the other. In the latter R expression, `which` is actually processing the values returned by the call to `is.na`. V... | s <- c(1, 4, 2, 3, 8)
sample(s)
sample(1:10, size = 5)
sample(0:1, size = 10, replace = TRUE) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
In summary, `sample(1:n, size = n)` returns a permutation of the `n` elements, while `sample(1:n, size = n, replace = TRUE)` provides a bootstrap sample of the original data. SortingSorting a list of values or finding the index or rank of any value in a vector are common tasks in statistical programming. It is differe... | z <- c(1, 6, 7, 2, 8, 3, 9, 4, 5)
sort(z)
order(z) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Data framesData frames are one of the core data structures to store and represent statistical data. Many routine functions that are used to load data stored in flat files or databases or to preprocess data stored in memory rely on data frames. Likewise, graphical commands such as those found in the `ggplot2` package g... | data(ToothGrowth)
head(ToothGrowth)
str(ToothGrowth) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
While `head` allows to preview the first 6 lines of a data frame, `str` provides a concise overview of what's available in the data frame, namely: the name of each variable (column), its mode of representation, et the first 10 observations (values).The dimensions (number of lines and columns) of a data frame can be ver... | dim(ToothGrowth) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
To access any given cell in this data frame, we will use the indexing trick we used in the case of vectors, but this time we have to indicate the line number as well as the column number, or name: Hence, `ToothGrowth[i,j]` means the value located at line `i` and column `j`, while `ToothGrowth[c(a,b),j]` would mean valu... | ToothGrowth[2,1] | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Since the columns of a data frame have names, it is equivalent to use `ToothGrowth[2,1]` and `ToothGrowth[2,"len"]`. In the latter case, variable names must be quoted. Column names can be displayed using `colnames` or `names` (in the special case of data frames), while row names are available *via* `rownames`. Row name... | ToothGrowth[c(2,4),1] | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
This amounts to 'indexed selection', meaning that we need to provide the row (or column) numbers, while most of the time we are interested in criterion-based indexation, that is: "which observation fullfills a given criterion." We generally call this a 'filter'. Since most R operations are vectorized, this happens to b... | head(ToothGrowth$supp[ToothGrowth$len > 6]) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
 Likewise, it is possible to combine different filters using logical operators: `&` stands for 'and' (logical conjunction) and `|` stands for 'or' (logical disjonction); logical equality is denoted as `==` while its negation reads `!=`. Here is an example where we want to select observati... | ToothGrowth[,ToothGrowth$len > 10 & ToothGrowth$dose < 1] | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
You will soon realize that for complex queries this notation become quite cumbersome: all variable must be prefixed by the name of the data frame, which can result in a very long statement. While it is recommended practice for programming or when developing dedicated package, it is easier to rely on `subset` in an inte... | subset(ToothGrowth, len > 10 & dose < 1) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
 It is also possible to use the technique discussed in the case of vectors to sort a data frame in ascending or descending order according to one or more variables. Here is an example using the `len` variable: | head(ToothGrowth)
head(ToothGrowth[order(ToothGrowth$len),]) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
The `which` function can also be used to retrieve a specific observation in a data frame, like in the following instruction: | which(ToothGrowth$len < 8) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Statistical summariesAs explained above, the `str` function is useful to check a given data structure, and individual properties of a data frame can be queried using dedicated functions, e.g. `nrow` or `ncol`. Now, to compute statistical quantities on a variable, we can use dedicated functions like `mean` (arithmetica... | mean(ToothGrowth$len)
range(ToothGrowth$len)
c(min(ToothGrowth$len), max(ToothGrowth$len))
table(ToothGrowth$dose)
summary(ToothGrowth) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Of course, the above functions can be applied to a subset of the original data set: | mean(ToothGrowth$len[ToothGrowth$dose == 1])
table(ToothGrowth$dose[ToothGrowth$len < 20]) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Bivariate caseIf we want to summarize a numerical variable according the values that a factor variable takes, we can use `tapply` or `aggregate`. The latter expects a 'formula' describing the relation between the variables we are interested in: the response variable or outcome appears on the left-hand side (LHS), whil... | aggregate(len ~ dose, data = ToothGrowth, mean)
aggregate(len ~ supp + dose, data = ToothGrowth, mean) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Note that only one function can be applied to the 'formula'. Even if it possible to write a custom function that computes the mean and standard deviation of a variable, both results will be returned as single column in the data frame returned by `aggregate`. There do exist other ways to perform such computation, though... | aggregate(len ~ dose, data = ToothGrowth, summary)
f <- function(x) c(mean = mean(x), sd = sd(x))
aggregate(len ~ dose, data = ToothGrowth, f) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
The `table` functions also works with two (or even three) variables: | table(ToothGrowth$dose, ToothGrowth$supp) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
If formulas are to be preferred, the `xtabs` function provides a convenient replacement for `table`: | xtabs(~ dose + supp, data = ToothGrowth) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
In either case, frequencies can be computed from the table of counts using `prop.table`, using the desired margin (row=1, column=2) in the bivariate case: | prop.table(table(ToothGrowth$dose))
prop.table(table(ToothGrowth$dose, ToothGrowth$supp), margins = 1) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Practical use case: The ESS surveyThe `data` directory includes three [RDS](https://www.rdocumentation.org/packages/base/versions/3.5.3/topics/readRDS) files related to the [European Social Survey](https://www.europeansocialsurvey.org) (ESS). This survey first ran in 2002 (round 1), and it is actually renewed every tw... | d <- readRDS("data/ess-one-round-fr.rds")
head(d[1:10])
table(d$yrbrn)
summary(d$agea) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Let us focus on the following list of variables, readily available in the file `ess-one-round-29vars-fr.rds`:- `tvtot`: TV watching, total time on average weekday- `rdtot`: Radio listening, total time on average weekday- `nwsptot`: Newspaper reading, total time on average weekday- `polintr`: How interested in politics-... | d <- readRDS("data/ess-one-round-29vars-fr.rds") | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
First, let us look at the distribution of the `gndr` variable, using a simple bar diagram: | summary(d$gndr)
p <- ggplot(data = d, aes(x = gndr)) +
geom_bar() +
labs(x = "Sex of respondant", y = "Counts")
p | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Now, let's look at the distribution of age. The `ggplot2` package offer a `geom_density` function but it is also possible to draw a line using the precomputed empirical density function, or to let `ggplot2` compute the density function itself using the `stat=` option. Here is how it looks: | summary(d$agea)
p <- ggplot(data = d, aes(x = agea)) +
geom_line(stat = "density", bw = 2) +
labs(x = "Age of respondant")
p | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
The distribution of age can also be represented as an histogram, and `ggplot2` makes it quite easy to split the display depending on the sex of the respondants, which is called a 'facet' in `ggplot2` parlance: | p <- ggplot(data = d, aes(x = agea)) +
geom_histogram(binwidth = 5) +
facet_grid(~ gndr) +
labs(x = "Age of respondant")
p | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Finally, a boxplot might also be an option, especially when we want to compare the distribution of a numerical variable across levels of a categorical variable. The `coord_flip` instruction is used to swap the X and Y axes, but keep in mind that `x=` and `y=` labels still refer to the `x=` and `y=` variable defined in ... | p <- ggplot(data = d, aes(x = gndr, y = agea)) +
geom_boxplot() +
coord_flip() +
labs(x = NULL, y = "Age of respondants")
p | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
**Sidenote:** In the above instructions, we used the following convention to build a `ggplot2` object:- we assign to a variable, say `p`, the call to `ggplot2` plus any further instructions ('geom', 'scale', 'coord_', etc.) using the `+` operator;- we use only one `aes()` structure, when calling `ggplot`, so that it m... | db <- readRDS("data/ess-one-round.rds")
cat("No. observations =", nrow(db))
table(db$cntry) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Since French data are (deliberately) missing from this dataset, we can append them to the above data frame as follows: | db <- rbind.data.frame(db, d)
cat("No. observations =", nrow(db))
db$cntry <- factor(db$cntry)
table(db$cntry) | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
Remember that is also possible to use `summary()` with a factor variable to display a table of counts. In this particular case, we are just appending a data frame to another data frame already loaded in memory. This assumes that both share the name columns, of course. Sometimes, another common operation might be perfor... | db$id <- 1:nrow(db)
db1 <- db[,c(1:10,ncol(db))]
db2 <- db[,c(11:(ncol(db)-1),ncol(db))]
all <- merge(db1, db2, by = "id") | _____no_output_____ | BSD-3-Clause | lang-r-base.ipynb | duchesnay/dspyr |
IMPORT | import os
import sys
import logging
import subprocess
import numpy as np
from shutil import copy
sys.path.insert(0, '/home/yongliang/third_party/merlin/src')
from io_funcs.binary_io import BinaryIOCollection
%load_ext autoreload
%autoreload 2 | The autoreload extension is already loaded. To reload it, use:
%reload_ext autoreload
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
UTILITY | def get_file_list_of_dir(dir_path):
res = [os.path.join(dir_path, f) for f in os.listdir(dir_path)]
res.sort()
return res
def gen_file_list(dir_path, file_id_list, ext):
return [os.path.join(dir_path, f + '.' + ext) for f in file_id_list]
def get_file_id_list(file_list):
return [os.path.splitext(o... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
CONFIGURATION | merlin_dir = '/home/yongliang/third_party/merlin'
silence_pattern = ['*-pau+*', '*-sil+*']
curr_dir = os.getcwd()
# hardcoded
nit_dir = os.path.join(curr_dir, 'nit')
wav_dir = os.path.join(nit_dir, 'wav2')
exp_dir = os.path.join(curr_dir, 'exp')
if not os.path.exists(exp_dir):
os.makedirs(exp_dir)
lab_dir = os.path... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Prepare label files | hmm_dir = os.path.join(exp_dir, 'hmm')
full_dir = os.path.join(nit_dir, 'full')
mono_dir = os.path.join(nit_dir, 'mono')
phones = os.path.join(nit_dir, 'monophone')
from src.forced_alignment import ForcedAlignment
aligner = ForcedAlignment(hmm_dir, wav_dir, full_dir, mono_dir, phones, lab_dir)
aligner.prepare_training... | ---make file_id_list.scp: /home/yongliang/third_party/merlin/egs/singing_synthesis/s3/exp/hmm/file_id_list.scp
---make copy.scp: /home/yongliang/third_party/merlin/egs/singing_synthesis/s3/exp/hmm/config/copy.scp
---mfcc extraction at: /home/yongliang/third_party/merlin/egs/singing_synthesis/s3/exp/hmm/mfc
------make c... | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Feature extraction | feat_dir = os.path.join(exp_dir, 'feat')
lf0_dir = os.path.join(feat_dir, 'lf0')
bap_dir = os.path.join(feat_dir, 'bap')
mgc_dir = os.path.join(feat_dir, 'mgc')
sample_rate = 16000
from src.feature_extraction import FeatureExtractor
feature_extractor = FeatureExtractor(wav_dir, sample_rate, feat_dir)
feature_extractor.... |
nitech_jp_song070_f001_003
Running REAPER f0 extraction...
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Duration model Model configuration | # duration model related
# hardcoded
dur_lab_dim = 368
dur_cmp_dim = 5
dur_train_file_number = 27
dur_valid_file_number = 1
dur_test_file_number = 1
dur_mdl_dir = os.path.join(exp_dir, 'duration_model')
if not os.path.exists(dur_mdl_dir):
os.makedirs(dur_mdl_dir)
dur_tmp_dir = os.path.join(dur_mdl_dir, 'tmp')
if... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Feature extraction from label files | ques_dir = os.path.join(curr_dir, 'ques')
question = os.path.join(ques_dir, 'general')
from frontend.label_normalisation import HTSLabelNormalisation
dur_lab_normaliser = HTSLabelNormalisation(question, add_frame_features=False, subphone_feats='none')
dur_lab_normaliser.perform_normalisation(orig_lab_file_list, dur_lab... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Remove silence phone | from frontend.silence_remover import SilenceRemover
dur_silence_remover = SilenceRemover(n_cmp=dur_lab_dim, silence_pattern=silence_pattern, remove_frame_features=False, subphone_feats='none')
dur_silence_remover.remove_silence(dur_lab_file_list, orig_lab_file_list, dur_lab_no_silence_file_list)
_, num_frame = io_funcs... | [[ 0. 0. 0. ... -1. -1. -1.]
[ 0. 0. 0. ... -1. -1. -1.]
[ 0. 0. 0. ... 192. 0. 100.]
...
[ 0. 0. 0. ... 72. 57. 43.]
[ 0. 0. 0. ... -1. -1. -1.]
[ 0. 0. 0. ... -1. -1. -1.]]
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Input feature normalization | from frontend.min_max_norm import MinMaxNormalisation
dur_min_max_normaliser = MinMaxNormalisation(feature_dimension=dur_lab_dim, min_value=0.01, max_value=0.99)
dur_min_max_normaliser.find_min_max_values(dur_lab_no_silence_file_list[0: dur_train_file_number])
dur_min_max_normaliser.normalise_data(dur_lab_no_silence_fi... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Compute duration from label files | dur_lab_normaliser.prepare_dur_data(orig_lab_file_list, dur_dur_file_list, feature_type='numerical')
feat, num_frame = io_funcs.load_binary_file_frame(dur_dur_file_list[0], 5)
print(feat.shape)
print(num_frame)
print(feat[0:10, :])
dur_dur_file_list[2] | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Make output features for duration model | delta_win = [-0.5, 0.0, 0.5]
acc_win = [1.0, -2.0, 1.0]
"""
"in" & "out" just mean before & after feature composition
like if we compute dynamic features, dimensions of out will be 3 times of in
not really mean in & out of the network
"""
dur_in_dimension_dict = {'dur': 5}
dur_out_dimension_dict = {'dur': 5}
dur_in... | (131, 5)
131
[[ 13. 72. 166. 2. 54.]
[ 9. 47. 1. 54. 32.]
[ 1. 18. 4. 4. 4.]
[ 3. 100. 1. 2. 6.]
[ 4. 5. 3. 3. 3.]
[ 1. 32. 1. 1. 3.]
[ 5. 3. 2. 2. 3.]
[ 2. 26. 1. 5. 8.]
[ 2. 3. 3. 7. 6.]
[ 6. 77. 17. 5. 5.]]
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Remove silence phone | dur_silence_remover = SilenceRemover(n_cmp = dur_cmp_dim, silence_pattern = silence_pattern, remove_frame_features = False, subphone_feats = 'none')
dur_silence_remover.remove_silence(dur_cmp_file_list, orig_lab_file_list, dur_cmp_no_silence_file_list)
_, num_frame = io_funcs.load_binary_file_frame(dur_cmp_file_list[2... | 131
124
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Output feature (duration) normalization | from frontend.mean_variance_norm import MeanVarianceNorm
dur_mvn_normaliser = MeanVarianceNorm(feature_dimension=dur_cmp_dim)
dur_global_mean_vector = dur_mvn_normaliser.compute_mean(dur_cmp_no_silence_file_list[0: dur_train_file_number], 0, dur_cmp_dim)
dur_global_std_vector = dur_mvn_normaliser.compute_std(dur_cmp_no... | [[ 4.77816306 32.451236 30.33884862 25.81974051 6.89332861]]
[[ 22.830841 1053.0828 920.4457 666.659 47.51798 ]]
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Model training | import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data as data
from torch.autograd import Variable
import math
import matplotlib.pyplot as plt
class DurationDataset(data.Dataset):
def __init__(self, lab_file_list, cmp_file_list, lab_dim=368, cmp_dim=5):
assert(len(lab_fi... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Test | print('files for test: ')
print(dur_lab_no_silence_norm_file_list[-1])
print(dur_cmp_no_silence_norm_file_list[-1])
print(dur_cmp_no_silence_file_list[-1])
print('*' * 20)
input_lab, num_input_frame = io_funcs.load_binary_file_frame(dur_lab_no_silence_norm_file_list[-1], 368)
target_cmp, num_target_frame = io_funcs... | files for test:
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Acoustic model Model configuration | # acoustic model related
# hardcoded
acou_lab_dim = 377
acou_cmp_dim = 187
acou_train_file_number = 27
acou_valid_file_number = 1
acou_test_file_number = 1
acou_mdl_dir = os.path.join(exp_dir, 'acoustic_model')
if not os.path.exists(acou_mdl_dir):
os.makedirs(acou_mdl_dir)
acou_inter_dir = os.path.join(acou_... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Feature extraction from label files | acou_lab_normaliser = HTSLabelNormalisation(question, add_frame_features=True, subphone_feats='full')
acou_lab_normaliser.perform_normalisation(orig_lab_file_list, acou_lab_file_list) | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Remove silence phone | acou_silence_remover = SilenceRemover(n_cmp=acou_lab_dim, silence_pattern=silence_pattern, remove_frame_features=True, subphone_feats='full')
acou_silence_remover.remove_silence(acou_lab_file_list, orig_lab_file_list, acou_lab_no_silence_file_list)
_, num_frame = io_funcs.load_binary_file_frame(acou_lab_file_list[2], 3... | [[0. 0. 0. ... 0.04234528 1. 0.00325733]
[0. 0. 0. ... 0.04234528 0.99674267 0.00651466]
[0. 0. 0. ... 0.04234528 0.99348533 0.00977199]
...
[0. 0. 0. ... 0.2925373 0.00895522 0.9940299 ]
[0. 0. ... | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Input feature normalization | acou_min_max_normaliser = MinMaxNormalisation(feature_dimension=acou_lab_dim, min_value=0.01, max_value=0.99)
acou_min_max_normaliser.find_min_max_values(acou_lab_no_silence_file_list[0: acou_train_file_number])
acou_min_max_normaliser.normalise_data(acou_lab_no_silence_file_list, acou_lab_no_silence_norm_file_list)
ac... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Make output features for acoustic model | """
"in" & "out" just mean before & after feature composition
like if we compute dynamic features, dimensions of out will be 3 times of in
not really mean in & out of the network
"""
acou_in_dimension_dict = {'bap': 1, 'mgc': 60, 'lf0': 1}
acou_out_dimension_dict = {'bap': 3, 'vuv': 1, 'mgc': 180, 'lf0': 3}
# acou_... | (131, 5)
[[ 13. 72. 166. 2. 54.]
[ 9. 47. 1. 54. 32.]
[ 1. 18. 4. 4. 4.]
[ 3. 100. 1. 2. 6.]
[ 4. 5. 3. 3. 3.]
[ 1. 32. 1. 1. 3.]
[ 5. 3. 2. 2. 3.]
[ 2. 26. 1. 5. 8.]
[ 2. 3. 3. 7. 6.]
[ 6. 77. 17. 5. 5.]]
(8640, 187)
[[ 0. 0. ... | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Remove silence phone | acou_silence_remover = SilenceRemover(n_cmp = acou_cmp_dim, silence_pattern = silence_pattern, remove_frame_features = True, subphone_feats = 'full')
acou_silence_remover.remove_silence(acou_cmp_file_list, orig_lab_file_list, acou_cmp_no_silence_file_list)
_, num_frame = io_funcs.load_binary_file_frame(acou_cmp_file_l... | 8640
7201
| Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Output feature (dim 187) normalization | acou_mvn_normaliser = MeanVarianceNorm(feature_dimension=acou_cmp_dim)
acou_global_mean_vector = acou_mvn_normaliser.compute_mean(acou_cmp_no_silence_file_list[0: acou_train_file_number], 0, acou_cmp_dim)
acou_global_std_vector = acou_mvn_normaliser.compute_std(acou_cmp_no_silence_file_list[0: acou_train_file_number], ... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Model training | # TODO to implement cross-validation, print something here to see whether could do normalisation in Pytorch
batch_size = int(acou_train_file_number)
# batch_size = 1
print('batch_size: ' + str(batch_size))
acou_train_set = DurationDataset(acou_lab_no_silence_norm_file_list[:10],
acou_cm... | _____no_output_____ | Apache-2.0 | egs/singing_synthesis/s3/run.ipynb | YongliangHe/SingingVoiceSynthesis |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.